Daily Papers

The rapid rise of Language Models (LMs) has expanded their use in several applications. Yet, due to constraints of model size, associated cost, or proprietary restrictions, utilizing state-of-the-art (SOTA) LLMs is not always feasible. With open, smaller LMs emerging, more applications can leverage their capabilities, but selecting the right LM can be challenging. This work conducts an in-depth experimental analysis of the semantic correctness of outputs of 10 smaller, open LMs across three aspects: task types, application domains and reasoning types, using diverse prompt styles. We demonstrate that most effective models and prompt styles vary depending on the specific requirements. Our analysis provides a comparative assessment of LMs and prompt styles using a proposed three-tier schema of aspects for their strategic selection based on use-case and other constraints. We also show that if utilized appropriately, these LMs can compete with, and sometimes outperform, SOTA LLMs like DeepSeek-v2, GPT-3.5-Turbo, and GPT-4o.

Multimodal visual language models are gaining prominence in open-world applications, driven by advancements in model architectures, training techniques, and high-quality data. However, their performance is often limited by insufficient task-specific data, leading to poor generalization and biased outputs. Existing efforts to increase task diversity in fine-tuning datasets are hindered by the labor-intensive process of manual task labeling, which typically produces only a few hundred task types. To address this, we propose TaskGalaxy, a large-scale multimodal instruction fine-tuning dataset comprising 19,227 hierarchical task types and 413,648 samples. TaskGalaxy utilizes GPT-4o to enrich task diversity by expanding from a small set of manually defined tasks, with CLIP and GPT-4o filtering those that best match open-source images, and generating relevant question-answer pairs. Multiple models are employed to ensure sample quality. This automated process enhances both task diversity and data quality, reducing manual intervention. Incorporating TaskGalaxy into LLaVA-v1.5 and InternVL-Chat-v1.0 models shows substantial performance improvements across 16 benchmarks, demonstrating the critical importance of task diversity. TaskGalaxy is publicly released at https://github.com/Kwai-YuanQi/TaskGalaxy.

In this paper, we explore the effects of language variants, data sizes, and fine-tuning task types in Arabic pre-trained language models. To do so, we build three pre-trained language models across three variants of Arabic: Modern Standard Arabic (MSA), dialectal Arabic, and classical Arabic, in addition to a fourth language model which is pre-trained on a mix of the three. We also examine the importance of pre-training data size by building additional models that are pre-trained on a scaled-down set of the MSA variant. We compare our different models to each other, as well as to eight publicly available models by fine-tuning them on five NLP tasks spanning 12 datasets. Our results suggest that the variant proximity of pre-training data to fine-tuning data is more important than the pre-training data size. We exploit this insight in defining an optimized system selection model for the studied tasks.

A person's demonstration often serves as a key reference for others learning the same task. However, RGB video, the dominant medium for representing these demonstrations, often fails to capture fine-grained contextual cues such as intent, safety-critical environmental factors, and subtle preferences embedded in human behavior. This sensory gap fundamentally limits the ability of Vision Language Models (VLMs) to reason about why actions occur and how they should adapt to individual users. To address this, we introduce MICA (Multimodal Interactive Contextualized Assistance), a framework that improves conversational agents for task assistance by integrating eye gaze and speech cues. MICA segments demonstrations into meaningful sub-tasks and extracts keyframes and captions that capture fine-grained intent and user-specific cues, enabling richer contextual grounding for visual question answering. Evaluations on questions derived from real-time chat-assisted task replication show that multimodal cues significantly improve response quality over frame-based retrieval. Notably, gaze cues alone achieves 93% of speech performance, and their combination yields the highest accuracy. Task type determines the effectiveness of implicit (gaze) vs. explicit (speech) cues, underscoring the need for adaptable multimodal models. These results highlight the limitations of frame-based context and demonstrate the value of multimodal signals for real-world AI task assistance.

The rapid proliferation of large language models (LLMs) and the growing diversity of their applications presents a unique optimization opportunity: selecting the right model for the task, while optimizing for speed, cost, and quality at a per-task level. However, inference endpoints can vary widely in quality, price, latency, context support, tool use, domain expertise, and reasoning behavior. This heterogeneity makes manual heuristics difficult to maintain and unlikely to achieve consistently favorable speed--cost--quality trade-offs on their own. We introduce , a lightweight GLiClass-based router that assigns a suitability score to each inference-time model label without autoregressive generation. The released 0.6B-parameter checkpoint combines a Qwen3 decoder with a shallow bidirectional scorer. Its decoder-KV execution path preserves a text-only key--value cache across a session, encodes only new dialogue turns, and evaluates transient candidate-label tokens without adding them to the persistent cache. The same checkpoint also predicts task type, difficulty, reasoning mode, and expected output length, and supports custom zero-shot labels. For task generation, we construct a task ontology with 23 families, 115 task types, 345 routable subtypes, 1,173 synthetic examples, and an orthogonal axis of 30 domains. Using this structure, we generate 150,000 verifier-scored tasks and 15,000 open-ended tasks. We then train the Qwen3 decoder on these tasks, while explicitly separating learned request prediction from per-task policies for attributes such as eligibility, cost, cache reuse, safety, and sovereignty. Across six LiveBench subsets, the router outperforms the mean candidate; on the selected 1,000-task subset, it achieves an aggregate top-1 score of 0.707 versus 0.696 for the strongest fixed model, with benchmark-dependent gains.

The advancement of general-purpose artificial intelligence relies on large language models (LLMs) that excel across a wide range of tasks, from structured reasoning to creative generation. However, post-training methods like Supervised Fine-Tuning (SFT) often struggle with generalization, favoring memorization over transferable learning. In this work, we introduce Omni-Thinker, a unified reinforcement learning (RL) framework that enhances LLM performance across diverse tasks by combining rule-based verifiable rewards with generative preference signals via LLM-as-a-Judge evaluations. Our approach enables consistent optimization across task types and scales RL-based training to subjective domains. We further investigate training strategies, demonstrating that a curriculum-based progression that orders tasks from structured to open-ended improves performance and reduces forgetting. Experimental results across four domains reveal that curriculum learning improves performance by 5.2% over joint training and 9.1% over model merging. These results highlight the importance of task-aware sampling and hybrid supervision in scaling RL-based post-training for general-purpose LLMs.

We introduce ClarQ-LLM, an evaluation framework consisting of bilingual English-Chinese conversation tasks, conversational agents and evaluation metrics, designed to serve as a strong benchmark for assessing agents' ability to ask clarification questions in task-oriented dialogues. The benchmark includes 31 different task types, each with 10 unique dialogue scenarios between information seeker and provider agents. The scenarios require the seeker to ask questions to resolve uncertainty and gather necessary information to complete tasks. Unlike traditional benchmarks that evaluate agents based on fixed dialogue content, ClarQ-LLM includes a provider conversational agent to replicate the original human provider in the benchmark. This allows both current and future seeker agents to test their ability to complete information gathering tasks through dialogue by directly interacting with our provider agent. In tests, LLAMA3.1 405B seeker agent managed a maximum success rate of only 60.05\%, showing that ClarQ-LLM presents a strong challenge for future research.

Comprehensive panoramic scene understanding is critical for immersive applications, yet it remains challenging due to the scarcity of high-resolution, multi-task annotations. While perspective foundation models have achieved success through data scaling, directly adapting them to the panoramic domain often fails due to severe geometric distortions and coordinate system discrepancies. Furthermore, the underlying relations between diverse dense prediction tasks in spherical spaces are underexplored. To address these challenges, we propose MTPano, a robust multi-task panoramic foundation model established by a label-free training pipeline. First, to circumvent data scarcity, we leverage powerful perspective dense priors. We project panoramic images into perspective patches to generate accurate, domain-gap-free pseudo-labels using off-the-shelf foundation models, which are then re-projected to serve as patch-wise supervision. Second, to tackle the interference between task types, we categorize tasks into rotation-invariant (e.g., depth, segmentation) and rotation-variant (e.g., surface normals) groups. We introduce the Panoramic Dual BridgeNet, which disentangles these feature streams via geometry-aware modulation layers that inject absolute position and ray direction priors. To handle the distortion from equirectangular projections (ERP), we incorporate ERP token mixers followed by a dual-branch BridgeNet for interactions with gradient truncation, facilitating beneficial cross-task information sharing while blocking conflicting gradients from incompatible task attributes. Additionally, we introduce auxiliary tasks (image gradient, point map, etc.) to fertilize the cross-task learning process. Extensive experiments demonstrate that MTPano achieves state-of-the-art performance on multiple benchmarks and delivers competitive results against task-specific panoramic specialist foundation models.

Prompt routing dynamically selects the most appropriate large language model from a pool of candidates for each query, optimizing performance while managing costs. As model pools scale to include dozens of frontier models with narrow performance gaps, existing approaches face significant challenges: manually defined task taxonomies cannot capture fine-grained capability distinctions, while monolithic routers struggle to differentiate subtle differences across diverse tasks. We propose a two-stage routing architecture that addresses these limitations through automated fine-grained task discovery and task-aware quality estimation. Our first stage employs graph-based clustering to discover latent task types and trains a classifier to assign prompts to discovered tasks. The second stage uses a mixture-of-experts architecture with task-specific prediction heads for specialized quality estimates. At inference, we aggregate predictions from both stages to balance task-level stability with prompt-specific adaptability. Evaluated on 10 benchmarks with 11 frontier models, our method consistently outperforms existing baselines and surpasses the strongest individual model while incurring less than half its cost.

In-context learning (ICL) enables Large Language Models (LLMs) to adapt to new tasks using few examples, with task vectors - specific hidden state activations - hypothesized to encode task information. Existing studies are limited by small-scale benchmarks, restricting comprehensive analysis. We introduce QuiteAFew, a novel dataset of 3,096 diverse few-shot tasks, each with 30 input-output pairs derived from the Alpaca dataset. Experiments with Llama-3-8B on QuiteAFew reveal: (1) task vector performance peaks at an intermediate layer (e.g., 15th), (2) effectiveness varies significantly by task type, and (3) complex tasks rely on multiple, subtask-specific vectors rather than a single vector, suggesting distributed task knowledge representation.

E-commerce agents contribute greatly to helping users complete their e-commerce needs. To promote further research and application of e-commerce agents, benchmarking frameworks are introduced for evaluating LLM agents in the e-commerce domain. Despite the progress, current benchmarks lack evaluating agents' capability to handle mixed-type e-commerce dialogue and complex domain rules. To address the issue, this work first introduces a novel corpus, termed Mix-ECom, which is constructed based on real-world customer-service dialogues with post-processing to remove user privacy and add CoT process. Specifically, Mix-ECom contains 4,799 samples with multiply dialogue types in each e-commerce dialogue, covering four dialogue types (QA, recommendation, task-oriented dialogue, and chit-chat), three e-commerce task types (pre-sales, logistics, after-sales), and 82 e-commerce rules. Furthermore, this work build baselines on Mix-Ecom and propose a dynamic framework to further improve the performance. Results show that current e-commerce agents lack sufficient capabilities to handle e-commerce dialogues, due to the hallucination cased by complex domain rules. The dataset will be publicly available.

Managing long sequences has become an important and necessary feature for large language models (LLMs). However, it is still an open question of how to comprehensively and systematically evaluate the long-sequence capability of LLMs. One of the reasons is that conventional and widely-used benchmarks mainly consist of short sequences. In this paper, we propose M4LE, a Multi-ability, Multi-range, Multi-task, Multi-domain benchmark for Long-context Evaluation. M4LE is based on a diverse NLP task pool comprising 36 NLP datasets, 11 task types and 12 domains. To alleviate the scarcity of tasks with naturally long sequences and incorporate multiple-ability assessment, we propose an automatic approach (but with negligible human annotations) to convert short-sequence tasks into a unified long-sequence scenario where LLMs have to identify single or multiple relevant spans in long contexts based on explicit or semantic hints. Specifically, the scenario includes five different types of abilities: (1) explicit single-span; (2) semantic single-span; (3) explicit multiple-span; (4) semantic multiple-span; and (5) global context understanding. The resulting samples in M4LE are evenly distributed from 1k to 8k input length. We conducted a systematic evaluation on 11 well-established LLMs, especially those optimized for long-sequence inputs. Our results reveal that: 1) Current LLMs struggle to understand long context, particularly when tasks require multiple-span attention. 2) Semantic retrieval task is more difficult for competent LLMs. 3) Models fine-tuned on longer text with position interpolation have comparable performance to those using Neural Tangent Kernel (NTK) aware scaling methods without fine-tuning. We make our benchmark publicly available to encourage future research in this challenging area.

We propose ADAPT, a meta-learning algorithm that learns task sampling proportions under an explicit token budget for multi-task instruction tuning. Instead of fixing task weights by hand, maintains a continuous distribution over tasks and updates it via meta-gradients of a smooth worst-case validation objective, inducing an adaptive curriculum that allocates more tokens to useful tasks while avoiding collapse. We instantiate ADAPT on three sim1B-parameter open-weight LLMs (Gemma-3-1B, LLaMA-3.2-1B, Qwen-0.6B), training on 20 Natural Instructions task types under budgets of 1%, 5%, and 10% of the available supervised tokens, and compare against strong supervised fine-tuning baselines with uniform and size-proportional mixing. We conduct evaluations on 11 out-of-domain benchmarks spanning reasoning, reading comprehension, code generation, and instruction following, we find that ADAPT matches or slightly improves average downstream performance relative to the best static mixture, while using fewer effective training tokens and reallocating budget toward harder, benchmark-aligned tasks.

Existing image perception methods based on VLMs generally follow a paradigm wherein models extract and analyze image content based on user-provided textual task prompts. However, such methods face limitations when applied to UAV imagery, which presents challenges like target confusion, scale variations, and complex backgrounds. These challenges arise because VLMs' understanding of image content depends on the semantic alignment between visual and textual tokens. When the task prompt is simplistic and the image content is complex, achieving effective alignment becomes difficult, limiting the model's ability to focus on task-relevant information. To address this issue, we introduce AerialVP, the first agent framework for task prompt enhancement in UAV image perception. AerialVP proactively extracts multi-dimensional auxiliary information from UAV images to enhance task prompts, overcoming the limitations of traditional VLM-based approaches. Specifically, the enhancement process includes three stages: (1) analyzing the task prompt to identify the task type and enhancement needs, (2) selecting appropriate tools from the tool repository, and (3) generating enhanced task prompts based on the analysis and selected tools. To evaluate AerialVP, we introduce AerialSense, a comprehensive benchmark for UAV image perception that includes Aerial Visual Reasoning, Aerial Visual Question Answering, and Aerial Visual Grounding tasks. AerialSense provides a standardized basis for evaluating model generalization and performance across diverse resolutions, lighting conditions, and both urban and natural scenes. Experimental results demonstrate that AerialVP significantly enhances task prompt guidance, leading to stable and substantial performance improvements in both open-source and proprietary VLMs. Our work will be available at https://github.com/lostwolves/AerialVP.

We introduce Bonito, an open-source model for conditional task generation: the task of converting unannotated text into task-specific training datasets for instruction tuning. Our goal is to enable zero-shot task adaptation of large language models on users' specialized, private data. We train Bonito on a new large-scale dataset with 1.65M examples created by remixing existing instruction tuning datasets into meta-templates. The meta-templates for a dataset produce training examples where the input is the unannotated text and the task attribute and the output consists of the instruction and the response. We use Bonito to generate synthetic tasks for seven datasets from specialized domains across three task types -- yes-no question answering, extractive question answering, and natural language inference -- and adapt language models. We show that Bonito significantly improves the average performance of pretrained and instruction tuned models over the de facto self supervised baseline. For example, adapting Mistral-Instruct-v2 and instruction tuned variants of Mistral and Llama2 with Bonito improves the strong zero-shot performance by 22.1 F1 points whereas the next word prediction objective undoes some of the benefits of instruction tuning and reduces the average performance by 0.8 F1 points. We conduct additional experiments with Bonito to understand the effects of the domain, the size of the training set, and the choice of alternative synthetic task generators. Overall, we show that learning with synthetic instruction tuning datasets is an effective way to adapt language models to new domains. The model, dataset, and code are available at https://github.com/BatsResearch/bonito.

With the integration of large language models (LLMs), embodied agents have strong capabilities to understand and plan complicated natural language instructions. However, a foreseeable issue is that those embodied agents can also flawlessly execute some hazardous tasks, potentially causing damages in the real world. Existing benchmarks predominantly overlook critical safety risks, focusing solely on planning performance, while a few evaluate LLMs' safety awareness only on non-interactive image-text data. To address this gap, we present SafeAgentBench-the first benchmark for safety-aware task planning of embodied LLM agents in interactive simulation environments. SafeAgentBench includes: (1) an executable, diverse, and high-quality dataset of 750 tasks, rigorously curated to cover 10 potential hazards and 3 task types; (2) SafeAgentEnv, a universal embodied environment with a low-level controller, supporting multi-agent execution with 17 high-level actions for 8 state-of-the-art baselines; and (3) reliable evaluation methods from both execution and semantic perspectives. Experimental results show that, although agents based on different design frameworks exhibit substantial differences in task success rates, their overall safety awareness remains weak. The most safety-conscious baseline achieves only a 10\% rejection rate for detailed hazardous tasks. Moreover, simply replacing the LLM driving the agent does not lead to notable improvements in safety awareness. More details and code are available at https://github.com/shengyin1224/SafeAgentBench.

Image-to-poster generation is a high-demand task requiring not only local adjustments but also high-level design understanding. Models must generate text, layout, style, and visual elements while preserving semantic fidelity and aesthetic coherence. The process spans two regimes: local editing, where ID-driven generation, rescaling, filling, and extending must preserve concrete visual entities; and global creation, where layout- and style-driven tasks rely on understanding abstract design concepts. These intertwined demands make image-to-poster a multi-dimensional process coupling entity-preserving editing with concept-driven creation under image-prompt control. To address these challenges, we propose PosterOmni, a generalized artistic poster creation framework that unlocks the potential of a base edit model for multi-task image-to-poster generation. PosterOmni integrates the two regimes, namely local editing and global creation, within a single system through an efficient data-distillation-reward pipeline: (i) constructing multi-scenario image-to-poster datasets covering six task types across entity-based and concept-based creation; (ii) distilling knowledge between local and global experts for supervised fine-tuning; and (iii) applying unified PosterOmni Reward Feedback to jointly align visual entity-preserving and aesthetic preference across all tasks. Additionally, we establish PosterOmni-Bench, a unified benchmark for evaluating both local editing and global creation. Extensive experiments show that PosterOmni significantly enhances reference adherence, global composition quality, and aesthetic harmony, outperforming all open-source baselines and even surpassing several proprietary systems.

Egocentric video understanding is inherently complex due to the dynamic 4D nature of the environment, where camera motion and object displacements necessitate a continuous re-evaluation of spatial relations. In this work, we target a suite of under-explored egocentric 4D reasoning tasks, including fixture interaction counting, viewpoint-relative fixture location, object movement itinerary tracking, and stationary object localization, that require fundamentally different cognitive operations: spatial anchoring, temporal tracking, and duration reasoning. We observe that these structural differences make task-agnostic approaches insufficient: generic Chain-of-Thought methods lack task-appropriate reasoning primitives, and uniform reinforcement learning actively destabilizes performance on spatial tasks. To address this, we propose EgoReasoner, a two-stage framework that aligns both the reasoning scaffold and the reward signal to each task's cognitive structure. In the first stage, Task-Adaptive Thinking Templates guide the synthesis of structured CoT traces that teach the model to reason adaptively across task types via supervised fine-tuning. In the second stage, task-aware reward functions verify entity grounding, temporal alignment, and task-adaptive logical consistency, selectively strengthening each reasoning pathway via reinforcement fine-tuning with GRPO. Our 3B-parameter model, trained on only 16K samples, achieves 37.5% average accuracy on the challenging HD-EPIC benchmark, surpassing Qwen2.5-VL-7B (25.7%) by over 10 points.

The evaluation of coding agents generally strives to be as realistic as possible. In our study, we collect 4,782 agent sessions of real software engineers in JetBrains IDEs, which we call Production Sessions. Since our subject is interactive agents, we study the sessions with at least three user messages (33% of the sample). These long sessions differ from issue-derived benchmark tasks in two ways: (i) user requests span a far wider mix of task types - questions about the project's code, planning, review, refactoring, execution - and (ii) users switch between types throughout a session. Long-session samples from three public interaction corpora exhibit markedly different Task Flows (the distributions of session lengths, task types, and type-to-type transitions), so no single interaction distribution is universally realistic: benchmarks should name a target use case and calibrate to measurements from it. We present SWE-TaskFlow, an approach for transforming any issue-derived benchmark: it preserves the verified tasks and tests while steering the interaction toward a target Task Flow through prompt splitting and verifiable repository QA, with a TaskFlow Alignment Score (TFAS) for selecting among generated trajectories. In a pilot on 700 SWE-Bench Pro tasks, solving the task sequentially in several steps approximately doubles agent cost without a stable change in resolve rate: the interaction protocol itself is an important dimension of evaluation.

Previous benchmarks on prompt injection in large language models (LLMs) have primarily focused on generic tasks and attacks, offering limited insights into more complex threats like data exfiltration. This paper examines how prompt injection can cause tool-calling agents to leak personal data observed during task execution. Using a fictitious banking agent, we develop data flow-based attacks and integrate them into AgentDojo, a recent benchmark for agentic security. To enhance its scope, we also create a richer synthetic dataset of human-AI banking conversations. In 16 user tasks from AgentDojo, LLMs show a 15-50 percentage point drop in utility under attack, with average attack success rates (ASR) around 20 percent; some defenses reduce ASR to zero. Most LLMs, even when successfully tricked by the attack, avoid leaking highly sensitive data like passwords, likely due to safety alignments, but they remain vulnerable to disclosing other personal data. The likelihood of password leakage increases when a password is requested along with one or two additional personal details. In an extended evaluation across 48 tasks, the average ASR is around 15 percent, with no built-in AgentDojo defense fully preventing leakage. Tasks involving data extraction or authorization workflows, which closely resemble the structure of exfiltration attacks, exhibit the highest ASRs, highlighting the interaction between task type, agent performance, and defense efficacy.

Visual Question-Answering (VQA) has become a key use-case in several applications to aid user experience, particularly after Vision-Language Models (VLMs) achieving good results in zero-shot inference. But evaluating different VLMs for an application requirement using a standardized framework in practical settings is still challenging. This paper introduces a comprehensive framework for evaluating VLMs tailored to VQA tasks in practical settings. We present a novel dataset derived from established VQA benchmarks, annotated with task types, application domains, and knowledge types, three key practical aspects on which tasks can vary. We also introduce GoEval, a multimodal evaluation metric developed using GPT-4o, achieving a correlation factor of 56.71% with human judgments. Our experiments with ten state-of-the-art VLMs reveals that no single model excelling universally, making appropriate selection a key design decision. Proprietary models such as Gemini-1.5-Pro and GPT-4o-mini generally outperform others, though open-source models like InternVL-2-8B and CogVLM-2-Llama-3-19B demonstrate competitive strengths in specific contexts, while providing additional advantages. This study guides the selection of VLMs based on specific task requirements and resource constraints, and can also be extended to other vision-language tasks.

We introduce SWE-Game, a benchmark of 247 tasks grounded in 41 executable reference Godot games spanning 13 gameplay categories in 2D and 3D. Five task types cover development from a brief, implementation from a game design document, skeleton completion, repair of 83 injected-fault cases, and Godot-to-Unity porting. Reference materials specify the intended gameplay, while a shared instrumentation interface lets evaluator-owned drivers and probes execute actions and observe independently implemented games. Evaluation combines engine-state checks, certified reference-input replay, and agent-authored feature demonstrations to assess mechanic correctness, demonstrated playability, and behavioral restoration and preservation after repairs. Game-specific vision-language rubrics separately assess presentation. Across six models, Opus5 achieves the highest overall score in all five task types. Best overall scores remain below 60 out of 100 across the three construction tasks, with Brief-to-Game reaching 50.38. Analysis of reviewed submissions identifies requirement omissions and gameplay logic errors as predominant implementation problems. On human-labeled behaviors from 100 agent-built games, executable checks achieve 92.59% balanced accuracy, compared with 78.41% for a video-based VLM judge. Rubric-based visual scores reach a Spearman correlation of 0.829 with human ratings of 200 gameplay clips. Together, these results characterize current agent capabilities across game-development activities and support combining runtime evidence with visual assessment.

The adoption of Large Language Models (LLMs) for code generation in data science offers substantial potential for enhancing tasks such as data manipulation, statistical analysis, and visualization. However, the effectiveness of these models in the data science domain remains underexplored. This paper presents a controlled experiment that empirically assesses the performance of four leading LLM-based AI assistants-Microsoft Copilot (GPT-4 Turbo), ChatGPT (o1-preview), Claude (3.5 Sonnet), and Perplexity Labs (Llama-3.1-70b-instruct)-on a diverse set of data science coding challenges sourced from the Stratacratch platform. Using the Goal-Question-Metric (GQM) approach, we evaluated each model's effectiveness across task types (Analytical, Algorithm, Visualization) and varying difficulty levels. Our findings reveal that all models exceeded a 50% baseline success rate, confirming their capability beyond random chance. Notably, only ChatGPT and Claude achieved success rates significantly above a 60% baseline, though none of the models reached a 70% threshold, indicating limitations in higher standards. ChatGPT demonstrated consistent performance across varying difficulty levels, while Claude's success rate fluctuated with task complexity. Hypothesis testing indicates that task type does not significantly impact success rate overall. For analytical tasks, efficiency analysis shows no significant differences in execution times, though ChatGPT tended to be slower and less predictable despite high success rates. This study provides a structured, empirical evaluation of LLMs in data science, delivering insights that support informed model selection tailored to specific task demands. Our findings establish a framework for future AI assessments, emphasizing the value of rigorous evaluation beyond basic accuracy measures.

We introduce POLLUX, a comprehensive open-source benchmark designed to evaluate the generative capabilities of large language models (LLMs) in Russian. Our main contribution is a novel evaluation methodology that enhances the interpretability of LLM assessment. For each task type, we define a set of detailed criteria and develop a scoring protocol where models evaluate responses and provide justifications for their ratings. This enables transparent, criteria-driven evaluation beyond traditional resource-consuming, side-by-side human comparisons. POLLUX includes a detailed, fine-grained taxonomy of 35 task types covering diverse generative domains such as code generation, creative writing, and practical assistant use cases, totaling 2,100 manually crafted and professionally authored prompts. Each task is categorized by difficulty (easy/medium/hard), with experts constructing the dataset entirely from scratch. We also release a family of LLM-as-a-Judge (7B and 32B) evaluators trained for nuanced assessment of generative outputs. This approach provides scalable, interpretable evaluation and annotation tools for model development, effectively replacing costly and less precise human judgments.

Transformers have emerged as the prevailing standard solution for various AI tasks, including computer vision and natural language processing. The widely adopted Query, Key, and Value formulation (QKV) has played a significant role in this. Nevertheless, no research has examined the essentiality of these three components for transformer performance. Therefore, we conducted an evaluation of the key-value formulation (KV), which generates symmetric attention maps, along with an asymmetric version that incorporates a 2D positional encoding into the attention matrix. Remarkably, this transformer requires fewer parameters and computation than the original one. Through experiments encompassing three task types -- synthetics (such as reversing or sorting a list), vision (mnist or cifar classification), and NLP (character generation and translation) -- we discovered that the KV transformer occasionally outperforms the QKV transformer. However, it also exhibits instances of underperformance compared to QKV, making it challenging to draw a definitive conclusion. Nonetheless, we consider the reported results to be encouraging and anticipate that they may pave the way for more efficient transformers in the future.

Recent advancements in foundation models have enhanced AI systems' capabilities in autonomous tool usage and reasoning. However, their ability in location or map-based reasoning - which improves daily life by optimizing navigation, facilitating resource discovery, and streamlining logistics - has not been systematically studied. To bridge this gap, we introduce MapEval, a benchmark designed to assess diverse and complex map-based user queries with geo-spatial reasoning. MapEval features three task types (textual, API-based, and visual) that require collecting world information via map tools, processing heterogeneous geo-spatial contexts (e.g., named entities, travel distances, user reviews or ratings, images), and compositional reasoning, which all state-of-the-art foundation models find challenging. Comprising 700 unique multiple-choice questions about locations across 180 cities and 54 countries, MapEval evaluates foundation models' ability to handle spatial relationships, map infographics, travel planning, and navigation challenges. Using MapEval, we conducted a comprehensive evaluation of 28 prominent foundation models. While no single model excelled across all tasks, Claude-3.5-Sonnet, GPT-4o, and Gemini-1.5-Pro achieved competitive performance overall. However, substantial performance gaps emerged, particularly in MapEval, where agents with Claude-3.5-Sonnet outperformed GPT-4o and Gemini-1.5-Pro by 16% and 21%, respectively, and the gaps became even more amplified when compared to open-source LLMs. Our detailed analyses provide insights into the strengths and weaknesses of current models, though all models still fall short of human performance by more than 20% on average, struggling with complex map images and rigorous geo-spatial reasoning. This gap highlights MapEval's critical role in advancing general-purpose foundation models with stronger geo-spatial understanding.

Multimodal Large Language Models (MLLMs) have shown significant progress in offline video understanding. However, applying these models to real-world scenarios, such as autonomous driving and human-computer interaction, presents unique challenges due to the need for real-time processing of continuous online video streams. To this end, this paper presents systematic efforts from three perspectives: evaluation benchmark, model architecture, and training strategy. First, we introduce OVBench, a comprehensive question-answering benchmark specifically designed to evaluate models' ability to perceive, memorize, and reason within online video contexts. It features six core task types across three temporal contexts-past, present, and future-forming 16 subtasks from diverse datasets. Second, we propose a new Pyramid Memory Bank (PMB) that effectively retains key spatiotemporal information in video streams. Third, we proposed an offline-to-online learning paradigm, designing an interleaved dialogue format for online video data and constructing an instruction-tuning dataset tailored for online video training. This framework led to the development of VideoChat-Online, a robust and efficient model for online video understanding. Despite the lower computational cost and higher efficiency, VideoChat-Online outperforms existing state-of-the-art offline and online models across popular offline video benchmarks and OVBench, demonstrating the effectiveness of our model architecture and training strategy.

The ability of critique is vital for models to self-improve and serve as reliable AI assistants. While extensively studied in language-only settings, multimodal critique of Large Multimodal Models (LMMs) remains underexplored despite their growing capabilities in tasks like captioning and visual reasoning. In this work, we introduce MM-CRITIC, a holistic benchmark for evaluating the critique ability of LMMs across multiple dimensions: basic, correction, and comparison. Covering 8 main task types and over 500 tasks, MM-CRITIC collects responses from various LMMs with different model sizes and is composed of 4471 samples. To enhance the evaluation reliability, we integrate expert-informed ground answers into scoring rubrics that guide GPT-4o in annotating responses and generating reference critiques, which serve as anchors for trustworthy judgments. Extensive experiments validate the effectiveness of MM-CRITIC and provide a comprehensive assessment of leading LMMs' critique capabilities under multiple dimensions. Further analysis reveals some key insights, including the correlation between response quality and critique, and varying critique difficulty across evaluation dimensions. Our code is available at https://github.com/MichealZeng0420/MM-Critic.

Evaluating large language models (LLMs) for software engineering has been limited by narrow task coverage, language bias, and insufficient alignment with real-world developer workflows. Existing benchmarks often focus on algorithmic problems or Python-centric bug fixing, leaving critical dimensions of software engineering underexplored. To address these gaps, we introduce SWE-Compass1, a comprehensive benchmark that unifies heterogeneous code-related evaluations into a structured and production-aligned framework. SWE-Compass spans 8 task types, 8 programming scenarios, and 10 programming languages, with 2000 high-quality instances curated from authentic GitHub pull requests and refined through systematic filtering and validation. We benchmark ten state-of-the-art LLMs under two agentic frameworks, SWE-Agent and Claude Code, revealing a clear hierarchy of difficulty across task types, languages, and scenarios. Moreover, by aligning evaluation with real-world developer practices, SWE-Compass provides a rigorous and reproducible foundation for diagnosing and advancing agentic coding capabilities in large language models.

We present Sphinx, a synthetic environment for visual perception and reasoning that targets core cognitive primitives. Sphinx procedurally generates puzzles using motifs, tiles, charts, icons, and geometric primitives, each paired with verifiable ground-truth solutions, enabling both precise evaluation and large-scale dataset construction. The benchmark covers 25 task types spanning symmetry detection, geometric transformations, spatial reasoning, chart interpretation, and sequence prediction. Evaluating recent large vision-language models (LVLMs) shows that even state-of-the-art GPT-5 attains only 51.1% accuracy, well below human performance. Finally, we demonstrate that reinforcement learning with verifiable rewards (RLVR) substantially improves model accuracy on these tasks and yields gains on external visual reasoning benchmarks, highlighting its promise for advancing multimodal reasoning.

Recent advances in video generation have been remarkable, enabling models to produce visually compelling videos with synchronized audio. While existing video generation benchmarks provide comprehensive metrics for visual quality, they lack convincing evaluations for audio-video generation, especially for models aiming to generate synchronized audio-video outputs. To address this gap, we introduce VABench, a comprehensive and multi-dimensional benchmark framework designed to systematically evaluate the capabilities of synchronous audio-video generation. VABench encompasses three primary task types: text-to-audio-video (T2AV), image-to-audio-video (I2AV), and stereo audio-video generation. It further establishes two major evaluation modules covering 15 dimensions. These dimensions specifically assess pairwise similarities (text-video, text-audio, video-audio), audio-video synchronization, lip-speech consistency, and carefully curated audio and video question-answering (QA) pairs, among others. Furthermore, VABench covers seven major content categories: animals, human sounds, music, environmental sounds, synchronous physical sounds, complex scenes, and virtual worlds. We provide a systematic analysis and visualization of the evaluation results, aiming to establish a new standard for assessing video generation models with synchronous audio capabilities and to promote the comprehensive advancement of the field.

Recent deep research agents primarily improve performance by scaling reasoning depth, but this leads to high inference cost and latency in search-intensive scenarios. Moreover, generalization across heterogeneous research settings remains challenging. In this work, we propose Search More, Think Less (SMTL), a framework for long-horizon agentic search that targets both efficiency and generalization. SMTL replaces sequential reasoning with parallel evidence acquisition, enabling efficient context management under constrained context budgets. To support generalization across task types, we further introduce a unified data synthesis pipeline that constructs search tasks spanning both deterministic question answering and open-ended research scenarios with task appropriate evaluation metrics. We train an end-to-end agent using supervised fine-tuning and reinforcement learning, achieving strong and often state of the art performance across benchmarks including BrowseComp (48.6\%), GAIA (75.7\%), Xbench (82.0\%), and DeepResearch Bench (45.9\%). Compared to Mirothinker-v1.0, SMTL with maximum 100 interaction steps reduces the average number of reasoning steps on BrowseComp by 70.7\%, while improving accuracy.

Large language models (LLMs) have shown impressive capabilities across diverse tasks, yet their performance on jurisdiction-specific legal reasoning remains underexplored. We present TW-LegalBench that utilizes Taiwanese legal system's rich official corpus open to the public to fill the gap in evaluating LLMs on Taiwanese law, among common-law benchmarks that focus on English sources and civil-law benchmarks focusing on sources of Simplified Chinese. TW-LegalBench comprises three task types: (1) over 16,000 multiple-choice questions (MCQs) across five years of official examinations in 18 professional domains; (2) 117 open-ended essay questions (OEQs) from examinations for legal professionals with official scoring rubrics; and (3) more than 14,000 legal judgment prediction (LJP) instances covering hundreds of crime categories. We evaluate 13 LLMs using accuracy for MCQs, a decomposed LLM-as-Judge framework based on the scoring rubric points for OEQs, and metrics for sentencing accuracy and statute citation for LJP. Our results reveal that top-performing models exceed the passing threshold for qualified lawyers (passing rate: 11%) but fall short of that for judges and prosecutors (passing rate: 1~2%). For LJP, while models demonstrate reasonable verdict type accuracy and sentence prediction capability, they struggle to cite exact legal articles. These findings highlight that reliable legal text generation remains challenging for LLMs, even though their performance on qualification examinations approaches human level.

A classical intuition holds that verifying a solution is easier than producing one. For today's coding agents, this intuition is being inverted: as foundation models develop stronger reasoning capabilities and engineering harnesses grow more sophisticated, generating complex candidate solutions is no longer difficult -- reliably verifying them has become the harder problem. Every verifier we can build is only a proxy for human intent, never the intent itself. This makes verification subject to a twofold difficulty: first, intent is underspecified by nature, making it inherently hard to faithfully check whether it has been fulfilled; second, during model training, optimization widens the gap between proxy and intent -- manifesting as reward hacking or signal saturation. To address this, we characterize the quality of verification signals along three dimensions -- scalability, faithfulness, and robustness -- and argue that achieving all three simultaneously is the central challenge. We further study four reward constructions: a test verifier for general coding tasks, a rubric verifier for frontend tasks, the user as verifier for real-world agent tasks, and an automated agent verifier for long-horizon tasks. Across different task types and policy capability levels, we conduct in-depth analysis and experiments on the core challenges of reward design and how to more effectively leverage reward signals. Experiments show that targeted verification design can effectively suppress reward hacking, improve task completion quality, and achieve significant gains across multiple internal and public benchmarks. These experiences collectively point to a core observation: no fixed reward function can remain effective as policy capability continues to grow; and verification must co-evolve with the generator.

Scientific ideas rarely start from a blank page. They inherit mechanisms, repair known limitations, and recombine pieces of earlier work, much like biological genomes. Current benchmarks still say little about whether AI systems can follow this inheritance structure. We present IdeaGene-Bench (IG-Bench), a benchmark for scientific lineage reasoning and lineage-grounded idea generation. IG-Bench is organized around the IdeaGene framework: each paper or proposal is represented as a set of minimal, typed, evidence-grounded Idea Genome objects, and a GenomeDiff aligns these objects to record inheritance, mutation, loss, external import, and novel insertion under six operational evolutionary dynamics. The benchmark contains 1,961 golden lineage traces, 1,085 curated Idea Genome objects, and 920 pairwise GenomeDiff records across 10 scientific domains. It supports two evaluations. IG-Exam (42 task types, 1,029 instances) tests closed-form lineage reasoning across Idea Genome abstraction, inheritance tracing, evolutionary reasoning, and lineage verification. IG-Arena evaluates generation with a lineage-conditioned Population-Evolution Score(PES), asking whether a proposal can be inserted as a coherent descendant of a given lineage population: it should inherit the right Idea Genome objects, vary meaningfully from nearby work, and offer selection value for future research. Experiments on 14 LLM-based scientists expose a compositional bottleneck. The strongest system reaches only 27.3% exact accuracy on lineage reasoning, and structured lineage context reshuffles system rankings rather than helping every participant uniformly.

Scientific research involves complex information-seeking and reasoning workflows across heterogeneous sources. However, existing benchmarks primarily emphasize general-domain retrieval or static scientific question answering, and therefore fail to assess key capabilities required in realistic scientific research workflows. We introduce SciExplore, a benchmark designed to evaluate scientific information-seeking and reasoning capabilities of LLMs and agents. SciExplore comprises four task types covering 103 expert-curated tasks across more than ten scientific disciplines: scientific database navigation, ambiguous literature retrieval, missing reference completion, and cross-source structured knowledge synthesis, which probe progressively higher-level abilities from entity-level reasoning and document-level identification to evidence-level grounding and domain-level synthesis. We evaluate over ten state-of-the-art LLMs and autonomous agents on SciExplore, revealing substantial performance gaps with performance degrading sharply as task complexity increases and extremely low accuracy on the most challenging structured synthesis tasks. These results highlight significant limitations of current models and agents in realistic scientific information-seeking scenarios.

A 3D CT scan entering a vision-language model produces a long sequence of visual tokens, often thousands to tens of thousands per volume, and this sequence must be compressed before a language model can consume it. Token compression is well studied in general vision, but little of it targets 3D CT specifically. A common baseline is grid average, which pools regular grid cells and can blend distinct anatomy, lesion, and air into one token. We present ORCA (ORgan-Centroid Aggregation), a token compressor for 3D CT. It merges adjacent tokens with organ guidance and adds a sinusoidal encoding of each region's centroid to preserve spatial layout. This preserves the anatomical information a downstream model needs. ORCA is training-free and plug-and-play, producing an adjustable token set without any model change or text query. We evaluate it across two datasets (CT-RATE and Merlin) and five encoders. The evaluation spans two task types: attribute prediction over five families (size, density, location, texture, and disease) and text generation (visual question answering and report generation). At matched token budgets, ORCA improves consistently over existing compression methods. It shrinks the visual context 64times and its KV-cache 50times, and is 31times faster to process each volume. Code released at https://github.com/renjie-liang/ORCA-3DCT.

Radiological artificial intelligence has advanced rapidly, yet most systems remain narrowly task-specific, data-intensive, and fragile under domain shift. Foundation models promise more transferable and data-efficient solutions, but existing approaches are limited in scale, evaluated narrowly, and often assume that a single pretrained model can support diverse downstream tasks. Here we present nnFoundation, complementary convolutional and transformer-based 3D radiological foundation models. Developed within the Human Radiome Project (THRP), nnFoundation is trained on 2.1 million CT, MRI, and PET image volumes from 125 institutional and public datasets. We evaluate them across 108 tasks spanning segmentation, detection, classification, report generation, and image retrieval, including evaluations under domain shift, by external partners and in low-data and low-compute regimes. Across all task types, our convolution- and transformer-based nnFoundation models consistently outperform both prior 3D foundation models and training from scratch, establishing state-of-the-art performance for radiological imaging. However, performance follows a consistent task-dependent structure: the convolutional nnFoundation model dominates spatially localized tasks, whereas the transformer-based nnFoundation model excels in tasks requiring global semantic reasoning and in frozen-feature settings. Dynamically aligning the foundation model topology with the dataset characteristics post-hoc further improves transfer across heterogeneous 3D settings. These results show that transferable 3D radiological performance is governed not by a single universal model, but by the interplay of scalable pretraining, complementary architectures, and dataset-aware adaptation. We release nnFoundation models integrated into nnU-Net and nnDetection, enabling immediate application across established radiology workflows.

Building datasets remains one of the most manual and brittle parts of AI development. In this technical report, we focus on the most extreme but also most prevalent setting real world practitioners face: a zero data regime. Here, practitioners don't have any data for the capability they want to learn. We introduce Invent-A-Dataset which is a prompt based system to go from dataset description to realistic and large scale post-training datasets. We evaluate Invent-A-Dataset against five frontier model APIs including Anthropic, Google, Open AI, DeepSeek, Zai. Across eight task types and dataset sizes up to 20K samples, Invent-A-Dataset significantly outperforms with both the highest quality (17% relative gains) while simultaneously producing the most diverse samples (19% relative gains). Its diversity advantage widens with scale of training dataset size (from parity at 200 samples to 37% relative gains at 20K samples). This translates into considerable downstream training gains, resulting in far more performant post-trained models. Invent-A-Dataset fine-tune consistently ranks higher compared to other generator fine-tunes across different post-trained model architectures.

The principal goal of the TREC Neural Cross-Language Information Retrieval (NeuCLIR) track is to study the effect of neural approaches on cross-language information access. The track has created test collections containing Chinese, Persian, and Russian news stories and Chinese academic abstracts. NeuCLIR includes four task types: Cross-Language Information Retrieval (CLIR) from news, Multilingual Information Retrieval (MLIR) from news, Report Generation from news, and CLIR from technical documents. A total of 274 runs were submitted by five participating teams (and as baselines by the track coordinators) for eight tasks across these four task types. Task descriptions and the available results are presented.

How well can NLP models generalize to a variety of unseen tasks when provided with task instructions? To address this question, we first introduce Super-NaturalInstructions, a benchmark of 1,616 diverse NLP tasks and their expert-written instructions. Our collection covers 76 distinct task types, including but not limited to classification, extraction, infilling, sequence tagging, text rewriting, and text composition. This large and diverse collection of tasks enables rigorous benchmarking of cross-task generalization under instructions -- training models to follow instructions on a subset of tasks and evaluating them on the remaining unseen ones. Furthermore, we build Tk-Instruct, a transformer model trained to follow a variety of in-context instructions (plain language task definitions or k-shot examples). Our experiments show that Tk-Instruct outperforms existing instruction-following models such as InstructGPT by over 9% on our benchmark despite being an order of magnitude smaller. We further analyze generalization as a function of various scaling parameters, such as the number of observed tasks, the number of instances per task, and model sizes. We hope our dataset and model facilitate future progress towards more general-purpose NLP models.

Although Large Language Models (LLMs) and Large Multimodal Models (LMMs) exhibit impressive skills in various domains, their ability for mathematical reasoning within visual contexts has not been formally examined. Equipping LLMs and LMMs with this capability is vital for general-purpose AI assistants and showcases promising potential in education, data analysis, and scientific discovery. To bridge this gap, we present MathVista, a benchmark designed to amalgamate challenges from diverse mathematical and visual tasks. We first taxonomize the key task types, reasoning skills, and visual contexts from the literature to guide our selection from 28 existing math-focused and visual question answering datasets. Then, we construct three new datasets, IQTest, FunctionQA, and PaperQA, to accommodate for missing types of visual contexts. The problems featured often require deep visual understanding beyond OCR or image captioning, and compositional reasoning with rich domain-specific tools, thus posing a notable challenge to existing models. We conduct a comprehensive evaluation of 11 prominent open-source and proprietary foundation models (LLMs, LLMs augmented with tools, and LMMs), and early experiments with GPT-4V. The best-performing model, Multimodal Bard, achieves only 58% of human performance (34.8% vs 60.3%), indicating ample room for further improvement. Given this significant gap, MathVista fuels future research in the development of general-purpose AI agents capable of tackling mathematically intensive and visually rich real-world tasks. Preliminary tests show that MathVista also presents challenges to GPT-4V, underscoring the benchmark's importance. The project is available at https://mathvista.github.io/.

Instruction tuned large language models (LLMs), such as ChatGPT, demonstrate remarkable performance in a wide range of tasks. Despite numerous recent studies that examine the performance of instruction-tuned LLMs on various NLP benchmarks, there remains a lack of comprehensive investigation into their ability to understand cross-lingual sociopragmatic meaning (SM), i.e., meaning embedded within social and interactive contexts. This deficiency arises partly from SM not being adequately represented in any of the existing benchmarks. To address this gap, we present SPARROW, an extensive multilingual benchmark specifically designed for SM understanding. SPARROW comprises 169 datasets covering 13 task types across six primary categories (e.g., anti-social language detection, emotion recognition). SPARROW datasets encompass 64 different languages originating from 12 language families representing 16 writing scripts. We evaluate the performance of various multilingual pretrained language models (e.g., mT5) and instruction-tuned LLMs (e.g., BLOOMZ, ChatGPT) on SPARROW through fine-tuning, zero-shot, and/or few-shot learning. Our comprehensive analysis reveals that existing open-source instruction tuned LLMs still struggle to understand SM across various languages, performing close to a random baseline in some cases. We also find that although ChatGPT outperforms many LLMs, it still falls behind task-specific finetuned models with a gap of 12.19 SPARROW score. Our benchmark is available at: https://github.com/UBC-NLP/SPARROW

Instruction-tuned large language models, such as T0, have demonstrated remarkable capabilities in following instructions across various domains. However, their proficiency remains notably deficient in many low-resource languages. To address this challenge, we introduce FarsInstruct: a comprehensive instruction dataset designed to enhance the instruction-following ability of large language models specifically for the Persian language, a significant yet underrepresented language globally. FarsInstruct encompasses a wide range of task types and datasets, each containing a mix of straightforward to complex manual written instructions, as well as translations from Public Pool of Prompts, ensuring a rich linguistic and cultural representation. Furthermore, we introduce Co-CoLA, a framework designed to enhance the multi-task adaptability of LoRA-tuned models. Through extensive experimental analyses, our study showcases the effectiveness of FarsInstruct dataset coupled with training by Co-CoLA framework, in improving the performance of large language models within the Persian context. As of the current writing, FarsInstruct comprises more than 200 templates across 21 distinct datasets, and we intend to update it consistently, thus augmenting its applicability.

Over 85 million computed tomography (CT) scans are performed annually in the US, of which approximately one quarter focus on the abdomen. Given the current radiologist shortage, there is a large impetus to use artificial intelligence to alleviate the burden of interpreting these complex imaging studies. Prior state-of-the-art approaches for automated medical image interpretation leverage vision language models (VLMs). However, current medical VLMs are generally limited to 2D images and short reports, and do not leverage electronic health record (EHR) data for supervision. We introduce Merlin - a 3D VLM that we train using paired CT scans (6+ million images from 15,331 CTs), EHR diagnosis codes (1.8+ million codes), and radiology reports (6+ million tokens). We evaluate Merlin on 6 task types and 752 individual tasks. The non-adapted (off-the-shelf) tasks include zero-shot findings classification (31 findings), phenotype classification (692 phenotypes), and zero-shot cross-modal retrieval (image to findings and image to impressions), while model adapted tasks include 5-year disease prediction (6 diseases), radiology report generation, and 3D semantic segmentation (20 organs). We perform internal validation on a test set of 5,137 CTs, and external validation on 7,000 clinical CTs and on two public CT datasets (VerSe, TotalSegmentator). Beyond these clinically-relevant evaluations, we assess the efficacy of various network architectures and training strategies to depict that Merlin has favorable performance to existing task-specific baselines. We derive data scaling laws to empirically assess training data needs for requisite downstream task performance. Furthermore, unlike conventional VLMs that require hundreds of GPUs for training, we perform all training on a single GPU.

Multimodal Large Language Models (MLLMs) are set to transform how machines process and generate human-like responses by integrating diverse modalities such as text, images, and code. Yet, effectively harnessing their capabilities hinges on optimal prompt engineering. We present a comprehensive experimental evaluation of seven prompt engineering methods applied to 13 open-source MLLMs over 24 tasks spanning Reasoning and Compositionality, Multimodal Understanding and Alignment, Complex Code Generation and Execution, and Knowledge Retrieval and Integration. Our approach stratifies models by parameter count into Small (<4B), Medium (4B-10B), and Large (>10B) categories and compares prompting techniques including Zero-Shot, One-Shot, Few-Shot, Chain-of-Thought, Analogical, Generated Knowledge, and Tree-of-Thought. While Large MLLMs excel in structured tasks such as code generation, achieving accuracies up to 96.88% under Few-Shot prompting, all models struggle with complex reasoning and abstract understanding, often yielding accuracies below 60% and high hallucination rates. Structured reasoning prompts frequently increased hallucination up to 75% in small models and led to longer response times (over 20 seconds in Large MLLMs), while simpler prompting methods provided more concise and efficient outputs. No single prompting method uniformly optimises all task types. Instead, adaptive strategies combining example-based guidance with selective structured reasoning are essential to enhance robustness, efficiency, and factual accuracy. Our findings offer practical recommendations for prompt engineering and support more reliable deployment of MLLMs across applications including AI-assisted coding, knowledge retrieval, and multimodal content understanding.

Visual personalization is essential in user-facing AI systems such as smart homes and healthcare, where aligning model behavior with user-centric concepts is critical. However, recent large Vision-Language Models (VLMs), despite their broad applicability, remain underexplored in their ability to adapt to individual users. In this paper, we introduce MMPB, the first extensive benchmark for evaluating VLMs on personalization. MMPB comprises 10k image-query pairs and includes 111 personalizable concepts across four categories: humans, animals, objects, and characters, with the human category enriched with preference-grounded queries. We structure personalization into three main task types, each highlighting a different key property of VLMs. Using 23 widely used VLMs including both open- and closed-source models, we evaluate personalization performance via a three-stage protocol: concept injection, multi-turn dialogue, and personalized querying. Our findings indicate that most VLMs (including some closed-source models) struggle with personalization, particularly in maintaining consistency over dialogue, handling user preferences, and adapting to visual cues. Our analysis reveals that the challenges in VLM personalization (such as refusal behaviors and long-context forgetting) highlight substantial room for improvement. By identifying these limitations and offering a scalable benchmark, MMPB offers valuable insights and a solid foundation for future research toward truly personalized multi-modal AI. Project Page: aidaslab.github.io/MMPB

With the widespread application of Large Language Models (LLMs) to various domains, concerns regarding the trustworthiness of LLMs in safety-critical scenarios have been raised, due to their unpredictable tendency to hallucinate and generate misinformation. Existing LLMs do not have an inherent functionality to provide the users with an uncertainty/confidence metric for each response it generates, making it difficult to evaluate trustworthiness. Although several studies aim to develop uncertainty quantification methods for LLMs, they have fundamental limitations, such as being restricted to classification tasks, requiring additional training and data, considering only lexical instead of semantic information, and being prompt-wise but not response-wise. A new framework is proposed in this paper to address these issues. Semantic density extracts uncertainty/confidence information for each response from a probability distribution perspective in semantic space. It has no restriction on task types and is "off-the-shelf" for new models and tasks. Experiments on seven state-of-the-art LLMs, including the latest Llama 3 and Mixtral-8x22B models, on four free-form question-answering benchmarks demonstrate the superior performance and robustness of semantic density compared to prior approaches.

We introduce EvaLearn, a pioneering benchmark designed to evaluate large language models (LLMs) on their learning capability and efficiency in challenging tasks, a critical, yet underexplored aspect of model potential. EvaLearn contains 648 challenging problems across six task types, grouped into 182 sequences, each sequence dedicated to one task type. Diverging from most existing benchmarks that evaluate models in parallel, EvaLearn requires models to solve problems sequentially, allowing them to leverage the experience gained from previous solutions. EvaLearn provides five comprehensive automated metrics to evaluate models and quantify their learning capability and efficiency. We extensively benchmark nine frontier models and observe varied performance profiles: some models, such as Claude-3.7-sonnet, start with moderate initial performance but exhibit strong learning ability, while some models struggle to benefit from experience and may even show negative transfer. Moreover, we investigate model performance under two learning settings and find that instance-level rubrics and teacher-model feedback further facilitate model learning. Importantly, we observe that current LLMs with stronger static abilities do not show a clear advantage in learning capability across all tasks, highlighting that EvaLearn evaluates a new dimension of model performance. We hope EvaLearn provides a novel evaluation perspective for assessing LLM potential and understanding the gap between models and human capabilities, promoting the development of deeper and more dynamic evaluation approaches. All datasets, the automatic evaluation framework, and the results studied in this paper are available at the GitHub repository.

Autonomous scientific research is significantly advanced thanks to the development of AI agents. One key step in this process is finding the right scientific literature, whether to explore existing knowledge for a research problem, or to acquire evidence for verifying assumptions and supporting claims. To assess AI agents' capability in driving this process, we present AutoResearchBench, a dedicated benchmark for autonomous scientific literature discovery. AutoResearchBench consists of two complementary task types: (1) Deep Research, which requires tracking down a specific target paper through a progressive, multi-step probing process, and (2) Wide Research, which requires comprehensively collecting a set of papers satisfying given conditions. Compared to previous benchmarks on agentic web browsing, AutoResearchBench is distinguished along three dimensions: it is research-oriented, calling for in-depth comprehension of scientific concepts; literature-focused, demanding fine-grained utilization of detailed information; and open-ended, involving an unknown number of qualified papers and thus requiring deliberate reasoning and search throughout. These properties make AutoResearchBench uniquely suited for evaluating autonomous research capabilities, and extraordinarily challenging. Even the most powerful LLMs, despite having largely conquered general agentic web-browsing benchmarks such as BrowseComp, achieve only 9.39% accuracy on Deep Research and 9.31% IoU on Wide Research, while many other strong baselines fall below 5%. We publicly release the dataset and evaluation pipeline to facilitate future research in this direction. We publicly release the dataset, evaluation pipeline, and code at https://github.com/CherYou/AutoResearchBench.

Using Multimodal Large Language Models (MLLMs) as judges to achieve precise and consistent evaluations has gradually become an emerging paradigm across various domains. Evaluating the capability and reliability of MLLM-as-a-judge systems is therefore essential for ensuring trustworthy assessment. Existing judge benchmarks categorize samples by task types but fail to capture the fundamental judgment capabilities required for reliable evaluation. In this work, we introduce M-JudgeBench, a ten-dimensional capability-oriented benchmark designed to comprehensively assess the judgment abilities of MLLMs. Our benchmark decomposes evaluation into pairwise Chain-of-Thought (CoT) comparison, length bias avoidance, and process error detection tasks, jointly covering ten fine-grained subtasks. This design enables diagnosis of model reliability across reasoning styles, response lengths, and cross-model variations. Systematic evaluation uncovers the systematic weaknesses in existing MLLM-as-a-judge systems. To address this issue, we further propose Judge-MCTS, a data construction framework generating pairwise reasoning trajectories with various correctness and length. Using Judge-MCTS, we construct an MCTS-augmented dataset and train M-Judger, a series of strong judge models. Extensive experiments demonstrate the superiority of M-Judger on existing judge benchmarks as well as M-JudgeBench. Overall, our work establishes a more principled foundation for evaluating MLLM-as-a-judge through M-JudgeBench and Judge-MCTS framework, paving the way for future research on judge model evaluation and capability-driven judge training.

Memory lets an embodied agent reuse past experience, yet retaining useful information does not ensure that the agent can apply it when conditions change. We present MemTransfer, a benchmark comparing six memory representations, a working-memory baseline and five representations of past experience, under a shared frozen vision-language-model policy. It comprises 100 navigation cases across ten task types in a simulated warehouse, with expert demonstrations supplying the history. Three comparisons vary the starting pose, route availability, and amount and task relevance of history. With one demonstration per task, Full-context and Episodic memory reach 95.3% and 100.0% success at the original demonstration start, but lose 48-49 percentage points at a new test start. Summary changes little between these two test starts, yet with four demonstrations per task it retains a smaller fraction of its unchanged-route success after blocking (39.3%) than Working memory (44.8%) or the two trajectory memories (56-58%). At the new test start, increasing from one to four relevant demonstrations raises Episodic success by 14.3 percentage points, while the other evaluated representations gain no more than 1.3 percentage points. Replacing half of the relevant histories with other-task experience lowers success for both trajectory memories. These results show that robustness to one kind of mismatch does not imply robustness to another, motivating evaluation of both stored information and its use at decision time.

Work agents operate over digital artifacts to execute professional knowledge-intensive work, requiring training environments that support long-horizon interaction and trustworthy verification. However, hand-crafted environments incur prohibitive engineering overhead that prevents environment scaling, whereas synthesis methods sacrifice workspace complexity, realism, or grounded verifiability. To bridge this gap, we introduce WorkForge, a scalable synthesis framework for constructing verifiable work-agent environments from real-world resources. Starting from expert workflows, WorkForge first identifies the resources, decisions, and deliverables required by each workflow. It then retrieves relevant real-world files and organizes them into a workspace. WorkForge inspects the workspace to extract concrete, checkable facts about its content. These factual anchors fix which task types the workspace can support and how their outcomes can be verified. Therefore, WorkForge derives each task's instructions, solution plan, and complementary programmatic and semantic verifiers directly from these factual anchors, keeping verification traceable to observable workspace evidence. Furthermore, we construct 16.7K verifiable environments across 40 professional domains, with workspaces collectively covering 60 file types. Post-training Qwen3.5-35B-A3B-Base improves GDPVal from 45.5 to 73.6 and APEX Score from 5.0 to 21.3, while enabling Qwen3.5-27B to achieve highly competitive performance and outperform strong competitors. Our analyses confirm the efficacy of the proposed method and reveal consistent scaling behaviors across both data volume and interaction horizons.

Large language models (LLMs) have demonstrated strong capabilities in various aspects. However, when applying them to the highly specialized, safe-critical legal domain, it is unclear how much legal knowledge they possess and whether they can reliably perform legal-related tasks. To address this gap, we propose a comprehensive evaluation benchmark LawBench. LawBench has been meticulously crafted to have precise assessment of the LLMs' legal capabilities from three cognitive levels: (1) Legal knowledge memorization: whether LLMs can memorize needed legal concepts, articles and facts; (2) Legal knowledge understanding: whether LLMs can comprehend entities, events and relationships within legal text; (3) Legal knowledge applying: whether LLMs can properly utilize their legal knowledge and make necessary reasoning steps to solve realistic legal tasks. LawBench contains 20 diverse tasks covering 5 task types: single-label classification (SLC), multi-label classification (MLC), regression, extraction and generation. We perform extensive evaluations of 51 LLMs on LawBench, including 20 multilingual LLMs, 22 Chinese-oriented LLMs and 9 legal specific LLMs. The results show that GPT-4 remains the best-performing LLM in the legal domain, surpassing the others by a significant margin. While fine-tuning LLMs on legal specific text brings certain improvements, we are still a long way from obtaining usable and reliable LLMs in legal tasks. All data, model predictions and evaluation code are released in https://github.com/open-compass/LawBench/. We hope this benchmark provides in-depth understanding of the LLMs' domain-specified capabilities and speed up the development of LLMs in the legal domain.

Reinforcement learning (RL) has shown promise for training multimodal agents to reason and use tools. However, it remains unclear when fine-grained credit assignment improves learning and when it fails to improve task performance. Video editing provides a useful testbed for this question, combining multimodal perception, long-horizon planning, and tool use with delayed outcome feedback. We construct four movie-trailer editing task types---Content Selection, Narrative Sequencing, Rough Cut, and Fine Cut---that reflect real-world editing workflows and span different levels of task complexity and interaction horizon. Decaying process rewards provides the largest benefit on Fine Cut, with a 45.4% relative gain over outcome-only GRPO on tasks averaging more than 60 turns under outcome-only GRPO. On the three shorter-horizon task types averaging 30--40 turns under the same baseline, outcome-only GRPO remains competitive, with decaying process rewards yielding at most a 10.4% relative gain. Process rewards from an LLM judge initially improve tool use and encourage visual inspection. As training progresses, however, GRPO+PRM with a fixed process-reward weight increasingly exhibits excessive reasoning and redundant tool calls: process rewards continue to rise while task completion rates and outcome scores decline. Decaying the process-reward weight to zero mitigates this reward-hacking behavior while preserving the early gains. Despite dense token-level supervision, on-policy self-distillation (OPSD) yields little improvement or substantial degradation and is highly sensitive to the choice of privileged information. Our analysis reveals differences in teacher--student disagreement and gradient magnitudes, and suggests possible dilution of informative signals over long trajectories.

Energy is now a critical ML computing resource. While measuring energy consumption and observing trends is a valuable first step, accurately understanding and diagnosing why those differences occur is crucial for optimization. To that end, we begin by presenting a large-scale measurement study of inference time and energy across the generative AI landscape with 46 models, 7 tasks, and 1,858 different configurations on NVIDIA H100 and B200 GPUs. Our empirical findings span order-of-magnitude variations: LLM task type can lead to 25times energy differences, video generation sometimes consumes more than 100times the energy of images, and GPU utilization differences can result in 3--5times energy differences. Based on our observations, we present a framework for reasoning about the underlying mechanisms that govern time and energy consumption. The essence is that time and energy are determined by latent metrics like memory and utilization, which are in turn affected by various factors across the algorithm, software, and hardware layers. Our framework also extends directly to throughput per watt, a critical metric for power-constrained datacenters.

Long-term conversational memory is a core capability for LLM-based dialogue systems, yet existing benchmarks and evaluation protocols primarily focus on surface-level factual recall. In realistic interactions, appropriate responses often depend on implicit constraints such as user state, goals, or values that are not explicitly queried later. To evaluate this setting, we introduce LoCoMo-Plus, a benchmark for assessing cognitive memory under cue--trigger semantic disconnect, where models must retain and apply latent constraints across long conversational contexts. We further show that conventional string-matching metrics and explicit task-type prompting are misaligned with such scenarios, and propose a unified evaluation framework based on constraint consistency. Experiments across diverse backbone models, retrieval-based methods, and memory systems demonstrate that cognitive memory remains challenging and reveals failures not captured by existing benchmarks. Our code and evaluation framework are publicly available at: https://github.com/xjtuleeyf/Locomo-Plus.

Measuring complexity from a model's own code is biased by failure: a hard prompt can produce a short, broken program that looks simple. We instead score each prompt's structure on six dimensions from the prompt alone, never from generated code or test results. We draw 5,000 Python prompts across six bands of a preliminary single-rater score, and four out-of-panel LLM judges rescore every prompt (composite ICC(2,1) = 0.872). Twenty-one models answer every prompt (105,000 generations), and an independent LLM auditor, checked against known-answer cases, reviews each unit-test verdict. Averaged over models, pass rate has a nonmonotone breakpoint at composite 14.0: 83.4% at or below, 90.1% above. That jump is not a law of difficulty: with task-type fixed effects the breakpoint stays at 14.0 but the adjusted jump shrinks from 6.8 to 0.4 points, and neither of the benchmark's two prompt sources reproduces it on its own. Nineteen models pass more often above their own breakpoint and two less often; one of those declines comes from API calls that returned nothing. In post hoc follow-up, Kimi K2.5 shows the clearest decline, at 14.25, under a prompt-source control and within the later source alone. Among failed generations with measurable output complexity, 25.0% pair a prompt composite above 8 with output complexity of at most 10, twice the share among passing ones. The pooled breakpoint moves to 11.25 on raw test outcomes. Human agreement with the index is moderate, paraphrase and cross-language rescoring preserve its ordering, and overidentification tests reject the six dimensions as instruments, so we make no causal claim. The contribution is a prompt-side way to locate where each model's pass rate changes and to test how much that depends on the prompt mix and on how success is scored.

Retrieval-Augmented Generation (RAG) has become a standard architectural pattern for incorporating domain-specific knowledge into user-facing chat applications powered by Large Language Models (LLMs). RAG systems are characterized by (1) a document retriever that queries a domain-specific corpus for context information relevant to an input query, and (2) an LLM that generates a response based on the provided query and context. However, comprehensive evaluation of RAG systems remains a challenge due to the lack of unified evaluation criteria and annotated datasets. In response, we introduce RAGBench: the first comprehensive, large-scale RAG benchmark dataset of 100k examples. It covers five unique industry-specific domains and various RAG task types. RAGBench examples are sourced from industry corpora such as user manuals, making it particularly relevant for industry applications. Further, we formalize the TRACe evaluation framework: a set of explainable and actionable RAG evaluation metrics applicable across all RAG domains. We release the labeled dataset at https://huggingface.co/datasets/rungalileo/ragbench. RAGBench explainable labels facilitate holistic evaluation of RAG systems, enabling actionable feedback for continuous improvement of production applications. Thorough extensive benchmarking, we find that LLM-based RAG evaluation methods struggle to compete with a finetuned RoBERTa model on the RAG evaluation task. We identify areas where existing approaches fall short and propose the adoption of RAGBench with TRACe towards advancing the state of RAG evaluation systems.

Foundation models pretrained on large-scale datasets are revolutionizing the field of computational pathology (CPath). The generalization ability of foundation models is crucial for the success in various downstream clinical tasks. However, current foundation models have only been evaluated on a limited type and number of tasks, leaving their generalization ability and overall performance unclear. To address this gap, we established a most comprehensive benchmark to evaluate the performance of off-the-shelf foundation models across six distinct clinical task types, encompassing a total of 39 specific tasks. Our findings reveal that existing foundation models excel at certain task types but struggle to effectively handle the full breadth of clinical tasks. To improve the generalization of pathology foundation models, we propose a unified knowledge distillation framework consisting of both expert and self knowledge distillation, where the former allows the model to learn from the knowledge of multiple expert models, while the latter leverages self-distillation to enable image representation learning via local-global alignment. Based on this framework, a Generalizable Pathology Foundation Model (GPFM) is pretrained on a large-scale dataset consisting of 190 million images from around 86,000 public H&E whole slides across 34 major tissue types. Evaluated on the established benchmark, GPFM achieves an impressive average rank of 1.36, with 29 tasks ranked 1st, while the the second-best model, UNI, attains an average rank of 2.96, with only 4 tasks ranked 1st. The superior generalization of GPFM demonstrates its exceptional modeling capabilities across a wide range of clinical tasks, positioning it as a new cornerstone for feature representation in CPath.

Variation in language is ubiquitous and often systematically linked to regional, social, and contextual factors. Tokenizers split texts into smaller units and might behave differently for less common linguistic forms. This might affect downstream LLM performance differently on two types of tasks: Tasks where the model should be robust to language variation (e.g., for semantic tasks like NLI, labels do not depend on whether a text uses British or American spelling) and tasks where the model should be sensitive to language variation (e.g., for form-based tasks like authorship verification, labels depend on whether a text uses British or American spelling). We pre-train BERT base models with the popular Byte-Pair Encoding algorithm to investigate how key tokenization design choices impact the performance of downstream models: the corpus used to train the tokenizer, the pre-tokenizer and the vocabulary size. We find that the best tokenizer varies on the two task types and that the pre-tokenizer has the biggest overall impact on performance. Further, we introduce a new approach to estimate tokenizer impact on downstream LLM performance, showing substantial improvement over metrics like R\'enyi efficiency. We encourage more work on language variation and its relation to tokenizers and thus LLM performance.

Large Language Models (LLMs) continue to advance natural language processing with their ability to generate human-like text across a range of tasks. Despite the remarkable success of LLMs in Natural Language Processing (NLP), their performance in text summarization across various domains and datasets has not been comprehensively evaluated. At the same time, the ability to summarize text effectively without relying on extensive training data has become a crucial bottleneck. To address these issues, we present a systematic evaluation of six LLMs across four datasets: CNN/Daily Mail and NewsRoom (news), SAMSum (dialog), and ArXiv (scientific). By leveraging prompt engineering techniques including zero-shot and in-context learning, our study evaluates the performance using the ROUGE and BERTScore metrics. In addition, a detailed analysis of inference times is conducted to better understand the trade-off between summarization quality and computational efficiency. For Long documents, introduce a sentence-based chunking strategy that enables LLMs with shorter context windows to summarize extended inputs in multiple stages. The findings reveal that while LLMs perform competitively on news and dialog tasks, their performance on long scientific documents improves significantly when aided by chunking strategies. In addition, notable performance variations were observed based on model parameters, dataset properties, and prompt design. These results offer actionable insights into how different LLMs behave across task types, contributing to ongoing research in efficient, instruction-based NLP systems.

Learning from long-horizon demonstrations with complex action sequences presents significant challenges for visual imitation learning, particularly in understanding temporal relationships of actions and spatial relationships between objects. In this paper, we propose a new agent framework that incorporates two dedicated reflection modules to enhance both plan and code generation. The plan generation module produces an initial action sequence, which is then verified by the plan reflection module to ensure temporal coherence and spatial alignment with the demonstration video. The code generation module translates the plan into executable code, while the code reflection module verifies and refines the generated code to ensure correctness and consistency with the generated plan. These two reflection modules jointly enable the agent to detect and correct errors in both the plan generation and code generation, improving performance in tasks with intricate temporal and spatial dependencies. To support systematic evaluation, we introduce LongVILBench, a benchmark comprising 300 human demonstrations with action sequences of up to 18 steps. LongVILBench emphasizes temporal and spatial complexity across multiple task types. Experimental results demonstrate that existing methods perform poorly on this benchmark, whereas our new framework establishes a strong baseline for long-horizon visual imitation learning.

User interface (UI) design is an iterative process in which designers progressively refine their work with design software such as Figma or Sketch. Recent advances in vision language models (VLMs) with tool invocation suggest these models can operate design software to edit a UI design through iteration. Understanding and enhancing this capacity is important, as it highlights VLMs' potential to collaborate with designers within conventional software. However, as no existing benchmark evaluates tool-based design performance, the capacity remains unknown. To address this, we introduce CANVAS, a benchmark for VLMs on tool-based user interface design. Our benchmark contains 598 tool-based design tasks paired with ground-truth references sampled from 3.3K mobile UI designs across 30 function-based categories (e.g., onboarding, messaging). In each task, a VLM updates the design step-by-step through context-based tool invocations (e.g., create a rectangle as a button background), linked to design software. Specifically, CANVAS incorporates two task types: (i) design replication evaluates the ability to reproduce a whole UI screen; (ii) design modification evaluates the ability to modify a specific part of an existing screen. Results suggest that leading models exhibit more strategic tool invocations, improving design quality. Furthermore, we identify common error patterns models exhibit, guiding future work in enhancing tool-based design capabilities.

Foundation models for medical imaging are typically pretrained on increasingly large datasets, following a "scale-at-all-costs" paradigm. However, this strategy faces two critical challenges: large-scale medical datasets often contain substantial redundancy and severe class imbalance that bias representation learning toward over-represented patterns, and indiscriminate training regardless of heterogeneity in data quality incurs considerable computational inefficiency. Here we demonstrate that active, principled data curation during pretraining can serve as a viable, cost-effective alternative to brute-force dataset enlargement. We introduce CheXficient, a chest X-ray (CXR) foundation model that selectively prioritizes informative training samples. CheXficient is pretrained on only 22.7% of 1,235,004 paired CXR images and reports while consuming under 27.3% of the total compute budget, yet achieving comparable or superior performance to its full-data counterpart and other large-scale pretrained models. We assess CheXficient across 20 individual benchmarks spanning 5 task types, including non-adapted off-the-shelf evaluations (zero-shot findings classification and crossmodal retrieval) and adapted downstream tasks (disease prediction, semantic segmentation, and radiology report generation). Further analyses show that CheXficient systematically prioritizes under-represented training samples, improving generalizability on long-tailed or rare conditions. Overall, our work offers practical insights into the data and computation demands for efficient pretraining and downstream adaptation of medical vision-language foundation models.

Large language models are rapidly evolving into interactive coding agents capable of end-to-end web coding, yet existing benchmarks evaluate only narrow slices of this capability, typically text-conditioned generation with static-correctness metrics, leaving visual fidelity, interaction quality, and codebase-level reasoning largely unmeasured. We introduce WebCompass, a multimodal benchmark that provides unified lifecycle evaluation of web engineering capability. Recognizing that real-world web coding is an iterative cycle of generation, editing, and repair, WebCompass spans three input modalities (text, image, video) and three task types (generation, editing, repair), yielding seven task categories that mirror professional workflows. Through a multi-stage, human-in-the-loop pipeline, we curate instances covering 15 generation domains, 16 editing operation types, and 11 repair defect types, each annotated at Easy/Medium/Hard levels. For evaluation, we adopt a checklist-guided LLM-as-a-Judge protocol for editing and repair, and propose a novel Agent-as-a-Judge paradigm for generation that autonomously executes generated websites in a real browser, explores interactive behaviors via the Model Context Protocol (MCP), and iteratively synthesizes targeted test cases, closely approximating human acceptance testing. We evaluate representative closed-source and open-source models and observe that: (1) closed-source models remain substantially stronger and more balanced; (2) editing and repair exhibit distinct difficulty profiles, with repair preserving interactivity better but remaining execution-challenging; (3) aesthetics is the most persistent bottleneck, especially for open-source models; and (4) framework choice materially affects outcomes, with Vue consistently challenging while React and Vanilla/HTML perform more strongly depending on task type.

The current electroencephalogram (EEG) based deep learning models are typically designed for specific datasets and applications in brain-computer interaction (BCI), limiting the scale of the models and thus diminishing their perceptual capabilities and generalizability. Recently, Large Language Models (LLMs) have achieved unprecedented success in text processing, prompting us to explore the capabilities of Large EEG Models (LEMs). We hope that LEMs can break through the limitations of different task types of EEG datasets, and obtain universal perceptual capabilities of EEG signals through unsupervised pre-training. Then the models can be fine-tuned for different downstream tasks. However, compared to text data, the volume of EEG datasets is generally small and the format varies widely. For example, there can be mismatched numbers of electrodes, unequal length data samples, varied task designs, and low signal-to-noise ratio. To overcome these challenges, we propose a unified foundation model for EEG called Large Brain Model (LaBraM). LaBraM enables cross-dataset learning by segmenting the EEG signals into EEG channel patches. Vector-quantized neural spectrum prediction is used to train a semantically rich neural tokenizer that encodes continuous raw EEG channel patches into compact neural codes. We then pre-train neural Transformers by predicting the original neural codes for the masked EEG channel patches. The LaBraMs were pre-trained on about 2,500 hours of various types of EEG signals from around 20 datasets and validated on multiple different types of downstream tasks. Experiments on abnormal detection, event type classification, emotion recognition, and gait prediction show that our LaBraM outperforms all compared SOTA methods in their respective fields. Our code is available at https://github.com/935963004/LaBraM.

We propose a lightweight and single-pass uncertainty quantification method for detecting hallucinations in Large Language Models. The method uses attention matrices to estimate uncertainty without requiring repeated sampling or external models. Specifically, we measure the Kullback-Leibler divergence between each attention head's distribution and a uniform reference distribution, and use these features in a logistic regression probe. Across multiple datasets, task types, and model families, attention divergence is highly predictive of answer correctness and performs competitively with existing uncertainty estimation methods. We find that this signal is concentrated in middle layers and on factual tokens such as named entities and numbers, suggesting that attention dynamics provides an efficient and interpretable white-box signal of model uncertainty.

With the development and widespread application of large language models (LLMs), the new paradigm of "Model as Product" is rapidly evolving, and demands higher capabilities to address complex user needs, often requiring precise workflow execution which involves the accurate understanding of multiple tasks. However, existing benchmarks focusing on single-task environments with limited constraints lack the complexity required to fully reflect real-world scenarios. To bridge this gap, we present the Extremely Complex Instruction Following Benchmark (EIFBENCH), meticulously crafted to facilitate a more realistic and robust evaluation of LLMs. EIFBENCH not only includes multi-task scenarios that enable comprehensive assessment across diverse task types concurrently, but also integrates a variety of constraints, replicating complex operational environments. Furthermore, we propose the Segment Policy Optimization (SegPO) algorithm to enhance the LLM's ability to accurately fulfill multi-task workflow. Evaluations on EIFBENCH have unveiled considerable performance discrepancies in existing LLMs when challenged with these extremely complex instructions. This finding underscores the necessity for ongoing optimization to navigate the intricate challenges posed by LLM applications.

Establishing a clear link between model predictions and the visual evidence that supports them is critical for transparency and reliability in multimodal reasoning, yet current multimodal large language model (MLLM) evaluations do not explicitly enforce this alignment. Existing benchmarks assess either textual answer correctness or pixel-level localization in isolation, leaving the coupling of reasoning and grounding an open challenge. We introduce VISTAQA, a comprehensive benchmark for joint evaluation of free-form answer correctness and pixel-level evidence grounding in visual question answering. VISTAQA comprises 1,157 expert-curated samples spanning six task types and six visual domains, ranging from direct perception to compositional and relational reasoning. VISTAQA requires models to not only answer correctly, but to also provide precise segmentation masks that support their answers. It also includes hallucination-aware examples where no valid visual evidence exists. To support this enhanced evaluation, we introduce GROVE, a unified evaluation metric that enforces joint correctness by combining textual accuracy and grounding quality via a per-sample geometric mean, ensuring neither dimension can compensate for deficiencies in the other. Comprehensive experiments across grounding-aware models and hybrid pipelines with general-purpose MLLMs reveal that even the strongest systems achieve limited performance under GROVE, highlighting a substantial gap between answer accuracy and visual evidence alignment.

In this paper, we introduce the Polish Massive Text Embedding Benchmark (PL-MTEB), a comprehensive benchmark for text embeddings in the Polish language. PL-MTEB comprises 30 diverse NLP tasks across five categories: classification, clustering, pair classification, information retrieval, and semantic text similarity. Within the scope of this work, we added 12 new Polish-language tasks to MTEB based on existing datasets and prepared two new datasets used to create four clustering tasks. We evaluated 30 publicly available text embedding models, including Polish and multilingual models. We analyzed the results in detail for specific task types and model sizes. We made the prepared datasets, the source code for evaluation, and the obtained results available to the public at https://github.com/rafalposwiata/pl-mteb.

Generative image editing models struggle with structured statistical charts when data modifications require geometric synchronization. We formalize this task as Visuo-Logical Cascading Editing (VLCE). However, existing methods remain confined to localized text substitutions and struggle with dependency-aware cascading updates. To systematically evaluate this capability, we introduce ChartSync, an expert-validated benchmark constructed via a programmatic rendering pipeline that guarantees deterministic visuo-logical coupling for the ground truth. ChartSync comprises 870 triplets across 9 chart categories and 4 task types, including 235 geometry-coupled VLCE instances that specifically test cascading text-to-geometry synchronization. We further evaluate these instances via a two-tier framework combining objective visual metrics with a vision-language model judge paradigm to assess low-level fidelity alongside multimodal comprehension and reasoning. Evaluating 14 image editing models and one code-mediated pipeline reveals a nuanced capability gap: most open-source models suffer severe drops in geometric synchronization, while only two frontier proprietary models show emerging VLCE capability, with their residual errors mainly involving semantic isolation and background corruption. Our detailed error analysis deconstructs these failure paradigms to identify core meta-abilities for guiding future multimodal architectures. The ChartSync dataset and code are publicly released at https://github.com/kaka-yjk/ChartSyncCodebase.

Large Audio-Language Models (LALMs) have recently achieved strong performance across various audio-centric tasks. However, hallucination, where models generate responses that are semantically incorrect or acoustically unsupported, remains largely underexplored in the audio domain. Existing hallucination benchmarks mainly focus on text or vision, while the few audio-oriented studies are limited in scale, modality coverage, and diagnostic depth. We therefore introduce HalluAudio, the first large-scale benchmark for evaluating hallucinations across speech, environmental sound, and music. HalluAudio comprises over 5K human-verified QA pairs and spans diverse task types, including binary judgments, multi-choice reasoning, attribute verification, and open-ended QA. To systematically induce hallucinations, we design adversarial prompts and mixed-audio conditions. Beyond accuracy, our evaluation protocol measures hallucination rate, yes/no bias, error-type analysis, and refusal rate, enabling a fine-grained analysis of LALM failure modes. We benchmark a broad range of open-source and proprietary models, providing the first large-scale comparison across speech, sound, and music. Our results reveal significant deficiencies in acoustic grounding, temporal reasoning, and music attribute understanding, underscoring the need for reliable and robust LALMs.

Large language models are increasingly deployed as agents that plan over long horizons and act through external tools. Most agents select actions through unconstrained generation over an accumulating history, leaving implicit the procedural knowledge of what to do, in what order, and under which conditions. As trajectories lengthen, agents can lose track of their objectives, invoke tools out of order, and repeat unproductive actions. We introduce the Procedural Graph: just as a knowledge graph organizes factual knowledge into (entity, relation, entity) triplets for what-is questions, a Procedural Graph organizes procedural knowledge into (procedure, relation, procedure) triplets for what-to-do questions. At each decision step, the framework localizes the agent's active node, and a guidance model translates the surrounding subgraph into step-level situational guidance that biases the solver's next action without dictating it. The graph is self-evolving: an LLM refiner contrasts failed trajectories with successful ones and edits the graph's topology and attributes, committing edits that preserve or improve held-out validation performance while retaining rejected ones to discourage repetition. Starting from a minimal skeleton, the loop builds graphs that match or surpass hand-designed ones. It can also repair a flawed expert prior. Across multiple datasets, task types, and LLMs, the Procedural Graph delivers consistent gains over memory-based baselines, and self-evolution further improves performance without manual engineering.

Spatial reasoning depends not only on metric properties such as distance, angle, and shape, but also on topological relations that remain invariant under continuous deformation. Cognitive science identifies these relations as foundational to spatial understanding, yet foundation-model evaluations largely focus on metric or viewpoint-dependent relations. We introduce MindTopo, a benchmark of topological intuition across five properties grounded in cognitive science and formal topology: continuity, separation, order, enclosure, and knots. MindTopo evaluates each property at two cognitive levels. Reasoning asks a model to identify topological relations or infer how they change. Planning instantiates a foundation model as a closed-loop agent whose policy selects environment actions. MindTopo contains 11,030 instances across 13 procedurally generated task types with controllable difficulty. We benchmark 14 MLLMs and study agent configurations augmented with image and video generation, including 3 video generative models in planning settings. Every MLLM performs better on reasoning than on planning, and the best-performing model remains far below observed human performance. On Qwen3-VL-2B-Instruct, supervised fine-tuning and reinforcement learning improve reasoning more than planning. Generated observations retain local cues and reach plausible endpoints, but audited rollouts do not reliably follow environment dynamics or preserve topology across transitions. Our website is at https://mind-topo.github.io/

Large language model (LLM) agents are increasingly capable of supporting scientific workflows, but their capabilities across diverse research activities remain difficult to assess systematically. Existing benchmarks often focus on selected scientific capabilities or task types, limiting a comprehensive characterization of scientific agents' strengths and weaknesses in realistic research workflows. We introduce SCI-AGENT-EXAM, a workflow-oriented benchmark comprising 120 expert-curated tasks across 11 scientific fields and 7 research task types defined by the Scientific Workflow Task Taxonomy (SWTT). Each task requires both executable artifacts (e.g., code and scripts) and textual artifacts (e.g., reports and explanations). To evaluate the heterogeneous task outputs, we develop a hybrid evaluation protocol combining deterministic programmatic verification for executable artifacts with a calibrated, expert-validated LLM-based judge for textual artifacts. Beyond outcome assessment, we analyze progress and solution stability along execution trajectories to characterize how agents reach their final results. Experiments with 10 frontier LLM agents reveal substantial performance gaps: GPT-6 Astra achieves the highest overall score of 80.9, yet only 26.0% of its runs exceed the 0.95 success threshold. Further analyses uncover distinct capability profiles across scientific task types and substantial variation in execution efficiency and stability. Our code and data are available at https://github.com/SKYLENAGE-AI/Sci-Agent-Exam.

With the current surge in spatial reasoning explorations, researchers have made significant progress in understanding indoor scenes, but still struggle with diverse applications such as robotics and autonomous driving. This paper aims to advance all-scale spatial reasoning across diverse scenarios by tackling two key challenges: 1) the heavy reliance on indoor 3D scans and labor-intensive manual annotations for dataset curation; 2) the absence of effective all-scale scene modeling, which often leads to overfitting to individual scenes. In this paper, we introduce a holistic solution that integrates a structured spatial reasoning knowledge system, scale-aware modeling, and a progressive training paradigm, as the first attempt to broaden the all-scale spatial intelligence of MLLMs to the best of our knowledge. Using a task-specific, specialist-driven automated pipeline, we curate over 38K video scenes across 5 spatial scales to create SpaceVista-1M, a dataset comprising approximately 1M spatial QA pairs spanning 19 diverse task types. While specialist models can inject useful domain knowledge, they are not reliable for evaluation. We then build an all-scale benchmark with precise annotations by manually recording, retrieving, and assembling video-based data. However, naive training with SpaceVista-1M often yields suboptimal results due to the potential knowledge conflict. Accordingly, we introduce SpaceVista-7B, a spatial reasoning model that accepts dense inputs beyond semantics and uses scale as an anchor for scale-aware experts and progressive rewards. Finally, extensive evaluations across 5 benchmarks, including our SpaceVista-Bench, demonstrate competitive performance, showcasing strong generalization across all scales and scenarios. Our dataset, model, and benchmark will be released on https://peiwensun2000.github.io/mm2km .

Spatial understanding of the physical world from 2D visual inputs hinges on two complementary forms of geometric knowledge: holistic 3D structural perception and fine-grained metric scale estimation. Existing multimodal large language models (MLLMs) typically address only one facet, ingesting either depth maps or point clouds as additional model inputs, which incurs substantial computational overhead and inherits the generalization limitations of upstream prediction models. We propose GAMSI, a dual-pathway Geometry-Aware MLLM for Spatial Intelligence that takes only RGB images as input while internalizing both forms of geometric prior within a unified autoregressive backbone. Specifically, we introduce Metric-Structure Decoupled Queries (MSDQ) which employ two groups of learnable queries to respectively extract dense metric signals and sparse structural cues from the shared visual context, with a task-decoupled attention mask further preventing the two pathways from contaminating each other. Building on this, an Expert-Guided Visual Grounding (EVG) module projects the aggregated cues back to frame-level visual features and aligns them with vision foundation models, which serve purely as training-time supervision, rather than as model inputs. We further build a multi-task spatial instruction-tuning dataset (MTS) comprising 152{,}776 samples spanning 13 task types and three visual modalities, consolidated from six public datasets. Trained with a two-stage curriculum, GAMSI achieves state-of-the-art performance on seven spatial intelligence benchmarks.

Multimodal Large Language Models (MLLMs) have demonstrated impressive performance in short video understanding. However, understanding long-form videos still remains challenging for MLLMs. This paper proposes TimeSuite, a collection of new designs to adapt the existing short-form video MLLMs for long video understanding, including a simple yet efficient framework to process long video sequence, a high-quality video dataset for grounded tuning of MLLMs, and a carefully-designed instruction tuning task to explicitly incorporate the grounding supervision in the traditional QA format. Specifically, based on VideoChat, we propose our long-video MLLM, coined as VideoChat-T, by implementing a token shuffling to compress long video tokens and introducing Temporal Adaptive Position Encoding (TAPE) to enhance the temporal awareness of visual representation. Meanwhile, we introduce the TimePro, a comprehensive grounding-centric instruction tuning dataset composed of 9 tasks and 349k high-quality grounded annotations. Notably, we design a new instruction tuning task type, called Temporal Grounded Caption, to peform detailed video descriptions with the corresponding time stamps prediction. This explicit temporal location prediction will guide MLLM to correctly attend on the visual content when generating description, and thus reduce the hallucination risk caused by the LLMs. Experimental results demonstrate that our TimeSuite provides a successful solution to enhance the long video understanding capability of short-form MLLM, achieving improvement of 5.6% and 6.8% on the benchmarks of Egoschema and VideoMME, respectively. In addition, VideoChat-T exhibits robust zero-shot temporal grounding capabilities, significantly outperforming the existing state-of-the-art MLLMs. After fine-tuning, it performs on par with the traditional supervised expert models.

The effectiveness of Large Language Models (LLMs) is heavily influenced by the reasoning strategies, or styles of thought, employed in their prompts. However, the interplay between these reasoning styles, model architecture, and task type remains poorly understood. To address this, we introduce StyleBench, a comprehensive benchmark for systematically evaluating reasoning styles across diverse tasks and models. We assess five representative reasoning styles, including Chain of Thought (CoT), Tree of Thought (ToT), Algorithm of Thought (AoT), Sketch of Thought (SoT), and Chain-of-Draft (CoD) on five reasoning tasks, using 15 open-source models from major families (LLaMA, Qwen, Mistral, Gemma, GPT-OSS, Phi, and DeepSeek) ranging from 270M to 120B parameters. Our large-scale analysis reveals that no single style is universally optimal. We demonstrate that strategy efficacy is highly contingent on both model scale and task type: search-based methods (AoT, ToT) excel in open-ended problems but require large-scale models, while concise styles (SoT, CoD) achieve radical efficiency gains on well-defined tasks. Furthermore, we identify key behavioral patterns: smaller models frequently fail to follow output instructions and default to guessing, while reasoning robustness emerges as a function of scale. Our findings offer a crucial roadmap for selecting optimal reasoning strategies based on specific constraints, we open source the benchmark in https://github.com/JamesJunyuGuo/Style_Bench.

The paper describes the open Russian medical language understanding benchmark covering several task types (classification, question answering, natural language inference, named entity recognition) on a number of novel text sets. Given the sensitive nature of the data in healthcare, such a benchmark partially closes the problem of Russian medical dataset absence. We prepare the unified format labeling, data split, and evaluation metrics for new tasks. The remaining tasks are from existing datasets with a few modifications. A single-number metric expresses a model's ability to cope with the benchmark. Moreover, we implement several baseline models, from simple ones to neural networks with transformer architecture, and release the code. Expectedly, the more advanced models yield better performance, but even a simple model is enough for a decent result in some tasks. Furthermore, for all tasks, we provide a human evaluation. Interestingly the models outperform humans in the large-scale classification tasks. However, the advantage of natural intelligence remains in the tasks requiring more knowledge and reasoning.

Introducing user-specified visual concepts in image editing is highly practical as these concepts convey the user's intent more precisely than text-based descriptions. We propose FreeEdit, a novel approach for achieving such reference-based image editing, which can accurately reproduce the visual concept from the reference image based on user-friendly language instructions. Our approach leverages the multi-modal instruction encoder to encode language instructions to guide the editing process. This implicit way of locating the editing area eliminates the need for manual editing masks. To enhance the reconstruction of reference details, we introduce the Decoupled Residual ReferAttention (DRRA) module. This module is designed to integrate fine-grained reference features extracted by a detail extractor into the image editing process in a residual way without interfering with the original self-attention. Given that existing datasets are unsuitable for reference-based image editing tasks, particularly due to the difficulty in constructing image triplets that include a reference image, we curate a high-quality dataset, FreeBench, using a newly developed twice-repainting scheme. FreeBench comprises the images before and after editing, detailed editing instructions, as well as a reference image that maintains the identity of the edited object, encompassing tasks such as object addition, replacement, and deletion. By conducting phased training on FreeBench followed by quality tuning, FreeEdit achieves high-quality zero-shot editing through convenient language instructions. We conduct extensive experiments to evaluate the effectiveness of FreeEdit across multiple task types, demonstrating its superiority over existing methods. The code will be available at: https://freeedit.github.io/.

Machine learning research has advanced in multiple aspects, including model structures and learning methods. The effort to automate such research, known as AutoML, has also made significant progress. However, this progress has largely focused on the architecture of neural networks, where it has relied on sophisticated expert-designed layers as building blocks---or similarly restrictive search spaces. Our goal is to show that AutoML can go further: it is possible today to automatically discover complete machine learning algorithms just using basic mathematical operations as building blocks. We demonstrate this by introducing a novel framework that significantly reduces human bias through a generic search space. Despite the vastness of this space, evolutionary search can still discover two-layer neural networks trained by backpropagation. These simple neural networks can then be surpassed by evolving directly on tasks of interest, e.g. CIFAR-10 variants, where modern techniques emerge in the top algorithms, such as bilinear interactions, normalized gradients, and weight averaging. Moreover, evolution adapts algorithms to different task types: e.g., dropout-like techniques appear when little data is available. We believe these preliminary successes in discovering machine learning algorithms from scratch indicate a promising new direction for the field.

As Large Language Models (LLMs) become increasingly integrated into our daily lives, the potential harms from deceptive behavior underlie the need for faithfully interpreting their decision-making. While traditional probing methods have shown some effectiveness, they remain best for narrowly scoped tasks while more comprehensive explanations are still necessary. To this end, we investigate meta-models-an architecture using a "meta-model" that takes activations from an "input-model" and answers natural language questions about the input-model's behaviors. We evaluate the meta-model's ability to generalize by training them on selected task types and assessing their out-of-distribution performance in deceptive scenarios. Our findings show that meta-models generalize well to out-of-distribution tasks and point towards opportunities for future research in this area. Our code is available at https://github.com/acostarelli/meta-models-public .

Multimodal Large Language Models (MLLMs) demonstrate robust zero-shot capabilities across diverse vision-language tasks after training on mega-scale datasets. However, dense prediction tasks, such as semantic segmentation and keypoint detection, pose significant challenges for MLLMs when represented solely as text outputs. Simultaneously, current MLLMs utilizing latent embeddings for visual task decoding generally demonstrate limited adaptability to both multi-task learning and multi-granularity scenarios. In this work, we present REF-VLM, an end-to-end framework for unified training of various visual decoding tasks. To address complex visual decoding scenarios, we introduce the Triplet-Based Referring Paradigm (TRP), which explicitly decouples three critical dimensions in visual decoding tasks through a triplet structure: concepts, decoding types, and targets. TRP employs symbolic delimiters to enforce structured representation learning, enhancing the parsability and interpretability of model outputs. Additionally, we construct Visual-Task Instruction Following Dataset (VTInstruct), a large-scale multi-task dataset containing over 100 million multimodal dialogue samples across 25 task types. Beyond text inputs and outputs, VT-Instruct incorporates various visual prompts such as point, box, scribble, and mask, and generates outputs composed of text and visual units like box, keypoint, depth and mask. The combination of different visual prompts and visual units generates a wide variety of task types, expanding the applicability of REF-VLM significantly. Both qualitative and quantitative experiments demonstrate that our REF-VLM outperforms other MLLMs across a variety of standard benchmarks. The code, dataset, and demo available at https://github.com/MacavityT/REF-VLM.

Large language models frequently rely on both contextual input and parametric knowledge to perform tasks. However, these sources can come into conflict, especially when retrieved documents contradict the model's parametric knowledge. We propose a diagnostic framework to systematically evaluate LLM behavior under context-memory conflict, where the contextual information diverges from their parametric beliefs. We construct diagnostic data that elicit these conflicts and analyze model performance across multiple task types. Our findings reveal that (1) knowledge conflict has minimal impact on tasks that do not require knowledge utilization, (2) model performance is consistently higher when contextual and parametric knowledge are aligned, (3) models are unable to fully suppress their internal knowledge even when instructed, and (4) providing rationales that explain the conflict increases reliance on contexts. These insights raise concerns about the validity of model-based evaluation and underscore the need to account for knowledge conflict in the deployment of LLMs.

OpenAI's ChatGPT Atlas introduces new capabilities for web interaction, enabling the model to analyze webpages, process user intents, and execute cursor and keyboard inputs directly within the browser. While its capacity for information retrieval tasks has been demonstrated, its performance in dynamic, interactive environments remains less explored. In this study, we conduct an early evaluation of Atlas's web interaction capabilities using browser-based games as test scenarios, including Google's T-Rex Runner, Sudoku, Flappy Bird, and Stein.world. We employ in-game performance scores as quantitative metrics to assess performance across different task types. Our results show that Atlas performs strongly in logical reasoning tasks like Sudoku, completing puzzles significantly faster than human baselines, but struggles substantially in real-time games requiring precise timing and motor control, often failing to progress beyond initial obstacles. These findings suggest that while Atlas demonstrates capable analytical processing, there remain notable limitations in dynamic web environments requiring real-time interaction. The website of our project can be found at https://atlas-game-eval.github.io.

Hyperparameter Optimization (HPO) is crucial to develop well-performing machine learning models. In order to ease prototyping and benchmarking of HPO methods, we propose carps, a benchmark framework for Comprehensive Automated Research Performance Studies allowing to evaluate N optimizers on M benchmark tasks. In this first release of carps, we focus on the four most important types of HPO task types: blackbox, multi-fidelity, multi-objective and multi-fidelity-multi-objective. With 3 336 tasks from 5 community benchmark collections and 28 variants of 9 optimizer families, we offer the biggest go-to library to date to evaluate and compare HPO methods. The carps framework relies on a purpose-built, lightweight interface, gluing together optimizers and benchmark tasks. It also features an analysis pipeline, facilitating the evaluation of optimizers on benchmarks. However, navigating a huge number of tasks while developing and comparing methods can be computationally infeasible. To address this, we obtain a subset of representative tasks by minimizing the star discrepancy of the subset, in the space spanned by the full set. As a result, we propose an initial subset of 10 to 30 diverse tasks for each task type, and include functionality to re-compute subsets as more benchmarks become available, enabling efficient evaluations. We also establish a first set of baseline results on these tasks as a measure for future comparisons. With carps (https://www.github.com/automl/CARP-S), we make an important step in the standardization of HPO evaluation.

Instruction tuning has become a foundation for unlocking the capabilities of large-scale pretrained models and improving their performance on complex tasks. Thus, the construction of high-quality instruction datasets is crucial for enhancing model performance and generalizability. Although current instruction datasets have reached tens of millions of samples, models finetuned on them may still struggle with complex instruction following and tasks in rare domains. This is primarily due to limited expansion in both ``coverage'' (coverage of task types and knowledge areas) and ``depth'' (instruction complexity) of the instruction set. To address this issue, we propose a systematic instruction data construction framework, which integrates a hierarchical labeling system, an informative seed selection algorithm, an evolutionary data synthesis process, and a model deficiency diagnosis with targeted data generation. These components form an iterative closed-loop to continuously enhance the coverage and depth of instruction data. Based on this framework, we construct InfinityInstruct-Subject, a high-quality dataset containing ~1.5 million instructions. Experiments on multiple foundation models and benchmark tasks demonstrate its effectiveness in improving instruction-following capabilities. Further analyses suggest that InfinityInstruct-Subject shows enlarged coverage and depth compared to comparable synthesized instruction datasets. Our work lays a theoretical and practical foundation for the efficient, continuous evolution of instruction datasets, moving from data quantity expansion to qualitative improvement.

Existing benchmarks for Large Language Model (LLM) agents focus on task completion under idealistic settings but overlook reliability in real-world, user-facing applications. In domains, such as in-car voice assistants, users often issue incomplete or ambiguous requests, creating intrinsic uncertainty that agents must manage through dialogue, tool use, and policy adherence. We introduce CAR-bench, a benchmark for evaluating consistency, uncertainty handling, and capability awareness in multi-turn, tool-using LLM agents in an in-car assistant domain. The environment features an LLM-simulated user, domain policies, and 58 interconnected tools spanning navigation, productivity, charging, and vehicle control. Beyond standard task completion, CAR-bench introduces Hallucination tasks that test agents' limit-awareness under missing tools or information, and Disambiguation tasks that require resolving uncertainty through clarification or internal information gathering. Baseline results reveal large gaps between occasional and consistent success on all task types. Even frontier reasoning LLMs achieve less than 50% consistent pass rate on Disambiguation tasks due to premature actions, and frequently violate policies or fabricate information to satisfy user requests in Hallucination tasks, underscoring the need for more reliable and self-aware LLM agents in real-world settings.

Agentic repository-level code understanding is essential for automating complex software engineering tasks, yet the field lacks reliable benchmarks. Existing evaluations often overlook the long tail topics and rely on popular repositories where Large Language Models (LLMs) can cheat via memorized knowledge. To address this, we introduce SWE-QA-Pro, a benchmark constructed from diverse, long-tail repositories with executable environments. We enforce topical balance via issue-driven clustering to cover under-represented task types and apply a rigorous difficulty calibration process: questions solvable by direct-answer baselines are filtered out. This results in a dataset where agentic workflows significantly outperform direct answering (e.g., a ~13-point gap for Claude Sonnet 4.5), confirming the necessity of agentic codebase exploration. Furthermore, to tackle the scarcity of training data for such complex behaviors, we propose a scalable synthetic data pipeline that powers a two-stage training recipe: Supervised Fine-Tuning (SFT) followed by Reinforcement Learning from AI Feedback (RLAIF). This approach allows small open models to learn efficient tool usage and reasoning. Empirically, a Qwen3-8B model trained with our recipe surpasses GPT-4o by 2.3 points on SWE-QA-Pro and substantially narrows the gap to state-of-the-art proprietary models, demonstrating both the validity of our evaluation and the effectiveness of our agentic training workflow.

Despite significant advancements in natural language generation, controlling language models to produce texts with desired attributes remains a formidable challenge. In this work, we introduce RSA-Control, a training-free controllable text generation framework grounded in pragmatics. RSA-Control directs the generation process by recursively reasoning between imaginary speakers and listeners, enhancing the likelihood that target attributes are correctly interpreted by listeners amidst distractors. Additionally, we introduce a self-adjustable rationality parameter, which allows for automatic adjustment of control strength based on context. Our experiments, conducted with two task types and two types of language models, demonstrate that RSA-Control achieves strong attribute control while maintaining language fluency and content consistency. Our code is available at https://github.com/Ewanwong/RSA-Control.

Model merging constructs versatile models by integrating task-specific models without requiring labeled data or expensive joint retraining. Although recent methods improve adaptability to heterogeneous tasks by generating customized merged models for each instance, they face two critical limitations. First, the instance-specific merged models lack reusability, restricting the exploitation of high-quality merging configurations and efficient batch inference. Second, these methods treat each task-specific model as a monolithic whole, overlooking the diverse mergeability of homologous components such as attention and multilayer perceptron layers, and the differing merging sensitivities across components. To address these limitations, we propose MERGE (Modular Expert Recombination for fine-Grained mErging), a method that enables component-wise model merging and input-aware, on-demand module recombination at inference. MERGE formulates component-wise merging as a bi-objective optimization problem that balances cross-task performance and storage efficiency, and develops a surrogate-assisted evolutionary algorithm to efficiently identify Pareto-optimal merging configurations. These high-quality configurations underpin a reusable modular expert library, from which a lightweight routing network dynamically activates and recombines modular experts to assemble input-specific models and enable efficient inference under storage constraints. Extensive experiments across various model scales, task types, and fine-tuning strategies demonstrate that MERGE consistently outperforms strong baselines and generalizes effectively.

In real-world applications, models are expected to perform reliably across diverse settings. Yet, many existing multimodal benchmarks expand task types without capturing the visual diversity needed to handle open-ended visual inputs. We present WorldBench, a challenging and visually diverse reasoning benchmark to evaluate Multimodal Large Language Models (MLLMs). We build a taxonomy of thousands of visual concepts across multiple domains (e.g., living things). Guided by this taxonomy, we curate a broad collection of images from search engines and existing datasets to comprehensively represent the visual world. Through structured trial-and-error, we manually design challenging questions that frontier MLLMs fail to answer. On quantitative and human evaluations, WorldBench achieves higher visual diversity than any existing diverse benchmark. Evaluating 15 MLLMs on WorldBench reveals weaknesses in visual understanding: even the strongest model reaches only 64.0% accuracy, while some models perform marginally above chance-level. We hope our work highlights the importance of visual diversity in building multimodal benchmarks.

Modern AI models achieve strong performance on many established benchmarks, yet they still fail on tasks that humans find almost trivial, such as manipulating a string or drawing a dog with five legs. These examples suggest that existing benchmarks may under-measure persistent blind spots in current systems. We introduce blind-spots-bench, a benchmark designed to expose such blind spots through tasks that appear simple for humans but remain challenging for modern AI. We collect raw questions from students in an AI course, clean and annotate them with structured reference solutions, and propose a task taxonomy tailored to the resulting dataset of 235 samples. We further develop an automated grading pipeline to evaluate a wide range of models, including open-weight and closed-source language, vision-language, and image-generation models. Our analysis on blind-spots-bench reveals that closed-source frontier models can substantially outperform open-weight models with even approx10% gap, even when they attain comparable performance on existing benchmarks. A more fine-grained analysis shows that no single model dominates across all task types, and that some tasks remain challenging for all evaluated models. These results highlight the value of blind-spots-bench as a diagnostic stress test for identifying concrete weaknesses in current modern models.

Clinical protocol-execution tasks -- checking a lab value, applying a threshold, placing a correctly structured FHIR order -- are natural candidates for RL from world feedback: once clinical SMEs encode decision logic into a verifier, that verifier grades unlimited rollouts without per-episode annotation. But applying RL requires a sound feedback channel and sufficient base capability. We audit MedAgentBench v1/v2, find a 41.7\% silent-finish ceiling that makes inaction the RL dominant strategy, and construct MedAgentBench-v3 (MAB-v3) (508 tasks, 8.9\% ceiling). Training Qwen3-8B exposes two structural barriers: a capability ceiling (10/20 task types have 0\% base performance, zero gradient) and a format-knowledge barrier (3/20 types require exact clinical codes undiscoverable by exploration). Pure RL reaches 18.2\% pass@1 vs.\ 34.1\% for rule-based SFT; the 15.9~pp gap is attributable entirely to these barriers. A decision/format-knowledge/lookup taxonomy predicts RL learnability and prescribes the fix: SFT to inject codes, RL to learn conditionals.

Large Multi-modality Models (LMMs) have made significant progress in visual understanding and generation, but still face challenges in visual editing, particularly in following complex instructions, preserving appearance consistency, and supporting flexible input formats. To study this gap, we introduce RISEBench, the first benchmark for evaluating Reasoning-Informed viSual Editing (RISE), and extend it to RISEBench++, a more comprehensive and fine-grained benchmark for this emerging task. RISEBench++ extends the taxonomy into a hierarchical scheme spanning six reasoning dimensions: Temporal, Causal, Spatial, Logical, and Counterfactual Reasoning, together with Hybrid Reasoning integrating multiple reasoning types across multi-turn edits. These dimensions are further decomposed into 12 subcategories and 65 fine-grained task types. We expand input formats to include multi-image conditioning and scale the benchmark to 1000 human-annotated test cases, released in English and Chinese. We also improve our evaluation framework, assessing Instruction Reasoning, Appearance Consistency, and Visual Plausibility with human judges and an LMM-as-a-judge approach for more reliable and calibrated judgements. Beyond benchmarking, we introduce RISE-Agent, a training-free agentic framework integrating reasoning-driven planning, tool-augmented execution, and verifier-guided refinement, outperforming most strong existing approaches across diverse RISE tasks. We evaluate 58 visual editing approaches, including 34 open-source models, 19 closed-source models, and 5 agentic methods. The results reveal substantial challenges in reasoning-based visual editing, with even the strongest evaluated approach, GPT-Image-2.5 Sunburst, achieving only 56.6% accuracy. RISEBench++ highlights the limitations of contemporary editing models, provides insights, and indicates future directions for reasoning-aware visual editing.

Despite progress in instruction-based video editing, unimodal textual instructions inherently struggle to convey fine-grained textures and complex dynamics. To bridge this perceptual gap, we propose Visual In-context Editing, a new paradigm elevating video editing from textual instructions to multi-modal visual guidance encompassing single image, image pair, and video pair. To facilitate this paradigm, we curate VicEdit-400K, the first large-scale dataset for visual in-context video editing. We develop an automated pipeline to generate 400K high-quality samples across ten task types, ensuring superior visual fidelity and semantic consistency through multi-dimensional filtering. Leveraging this foundation, we introduce VicEdit, a unified framework to bridge visual and textual contexts. To adaptively extract editing semantics from heterogeneous references, we design Modality-Adaptive Semantic Distillation, which produces modality-specific semantic tokens from visual references. These tokens are then synergistically integrated with textual instructions through Dual-Context Injection, enabling the generation process to benefit from both visual and textual signals. Extensive evaluations on VicEditBench demonstrate that VicEdit achieves state-of-the-art performance across both basic instruction editing and visual in-context editing tasks, establishing visual in-context learning as a powerful and controllable paradigm for video editing.

Recent advancements in Large Language Models (LLMs) have enabled sophisticated reasoning and content generation, yet their inherent stochasticity poses significant challenges for ensuring predictive credibility. While traditional uncertainty taxonomy paradigms, such as the dichotomy of aleatoric and epistemic uncertainties, provide conceptual foundations, they often fail to capture the multi-component and multi-stage nature of LLM generation and struggle to evaluate the effectiveness of various Uncertainty Quantification (UQ) methods. In this paper, we propose a granular uncertainty taxonomy that systematically attributes LLM uncertainty into input-level, parameter-level, token-level, and decoding-process sources. Correspondingly, we categorize existing UQ methods into Bayesian, ensemble, consensus-based, and single-pass approaches. Furthermore, we introduce a comprehensive evaluation framework covering diverse generation settings and metrics. We empirically evaluate 21 typical UQ methods across three prominent LLM families, including Qwen3, Llama 3.2, and DeepSeek-V3, on benchmarks such as TriviaQA, GSM8K, and HumanEval. Our experimental results demonstrate that (i) the effectiveness of UQ methods is sensitive to task types and generation settings; (ii) consensus-based methods, typed Deg and EigV, consistently outperform other UQ approaches; and (iii) larger model scales correlate with lower uncertainty estimates, suggesting an empirical scaling law for LLM uncertainty. This work bridges the gap between theoretical origins and practical deployment, providing a versatile diagnostic tool for systematically quantifying uncertainty in LLM applications.

Large Language Model (LLM)-based agents are increasingly used for software engineering tasks, yet their performance is not determined by the base model alone. The agent harness substantially shapes how SE agents interact with repositories, execute actions, and validate solutions. However, the role of harness design remains insufficiently understood, especially across different models, tasks, and harness components. In this paper, we present a systematic empirical study of harness effects in SE agents. We first evaluate two representative harnesses, mini-SWE-agent and OpenCode, with ten models from two prominent open-weight model families, Qwen and DeepSeek, on three benchmarks: SWE-bench Pro, ProgramBench, and GitTaskBench. We then construct NanoHarness, a lightweight modular harness built on top of mini-SWE-agent, and use it to analyze five representative harness components: tool registry, context compression, explicit planning, subagents, and lazy skills. Experimental results show that harness effectiveness depends jointly on model capability and task type. Complex harnesses provide diminishing marginal gains on SWE-style issue repair as model capability improves, but can benefit stronger models on more complex and open-ended repository-level tasks. Component-level analysis on ProgramBench further shows that structured tool use and task-specific subagents provide the most stable improvements, while context compression and general subagents can hurt repository-generation performance. When combined, NanoHarness improves over mini-SWE-agent by 7.37 and 6.21 percentage points on Qwen3.7-Max and DeepSeek-V4-Pro, respectively, recovering most of the gains of product-level harnesses. These findings highlight harness design as a first-class factor in SE-agent performance and provide insights for building more effective and efficient coding agents.

Visual Question Answering (VQA) and Image Captioning (CAP), which are among the most popular vision-language tasks, have analogous scene-text versions that require reasoning from the text in the image. Despite their obvious resemblance, the two are treated independently and, as we show, yield task-specific methods that can either see or read, but not both. In this work, we conduct an in-depth analysis of this phenomenon and propose UniTNT, a Unified Text-Non-Text approach, which grants existing multimodal architectures scene-text understanding capabilities. Specifically, we treat scene-text information as an additional modality, fusing it with any pretrained encoder-decoder-based architecture via designated modules. Thorough experiments reveal that UniTNT leads to the first single model that successfully handles both task types. Moreover, we show that scene-text understanding capabilities can boost vision-language models' performance on general VQA and CAP by up to 2.69% and 0.6 CIDEr, respectively.

Prior research works have evaluated quantized LLMs using limited metrics such as perplexity or a few basic knowledge tasks and old datasets. Additionally, recent large-scale models such as Llama 3.1 with up to 405B have not been thoroughly examined. This paper evaluates the performance of instruction-tuned LLMs across various quantization methods (GPTQ, AWQ, SmoothQuant, and FP8) on models ranging from 7B to 405B. Using 13 benchmarks, we assess performance across six task types: commonsense Q\&A, knowledge and language understanding, instruction following, hallucination detection, mathematics, and dialogue. Our key findings reveal that (1) quantizing a larger LLM to a similar size as a smaller FP16 LLM generally performs better across most benchmarks, except for hallucination detection and instruction following; (2) performance varies significantly with different quantization methods, model size, and bit-width, with weight-only methods often yielding better results in larger models; (3) task difficulty does not significantly impact accuracy degradation due to quantization; and (4) the MT-Bench evaluation method has limited discriminatory power among recent high-performing LLMs.

Long-form writing agents require flexible integration and interaction across information retrieval, reasoning, and composition. Current approaches rely on predetermined workflows and rigid thinking patterns to generate outlines before writing, resulting in constrained adaptability during writing. In this paper we propose a general agent framework that achieves human-like adaptive writing through recursive task decomposition and dynamic integration of three fundamental task types, i.e. retrieval, reasoning, and composition. Our methodology features: 1) a planning mechanism that interleaves recursive task decomposition and execution, eliminating artificial restrictions on writing workflow; and 2) integration of task types that facilitates heterogeneous task decomposition. Evaluations on both fiction writing and technical report generation show that our method consistently outperforms state-of-the-art approaches across all automatic evaluation metrics, which demonstrate the effectiveness and broad applicability of our proposed framework.

AI agents have the potential to significantly alter the cybersecurity landscape. Here, we introduce the first framework to capture offensive and defensive cyber-capabilities in evolving real-world systems. Instantiating this framework with BountyBench, we set up 25 systems with complex, real-world codebases. To capture the vulnerability lifecycle, we define three task types: Detect (detecting a new vulnerability), Exploit (exploiting a given vulnerability), and Patch (patching a given vulnerability). For Detect, we construct a new success indicator, which is general across vulnerability types and provides localized evaluation. We manually set up the environment for each system, including installing packages, setting up server(s), and hydrating database(s). We add 40 bug bounties, which are vulnerabilities with monetary awards from \10 to 30,485, covering 9 of the OWASP Top 10 Risks. To modulate task difficulty, we devise a new strategy based on information to guide detection, interpolating from identifying a zero day to exploiting a given vulnerability. We evaluate 10 agents: Claude Code, OpenAI Codex CLI with o3-high and o4-mini, and custom agents with o3-high, GPT-4.1, Gemini 2.5 Pro Preview, Claude 3.7 Sonnet Thinking, Qwen3 235B A22B, Llama 4 Maverick, and DeepSeek-R1. Given up to three attempts, the top-performing agents are Codex CLI: o3-high (12.5% on Detect, mapping to \3,720; 90% on Patch, mapping to 14,152), Custom Agent: Claude 3.7 Sonnet Thinking (67.5% on Exploit), and Codex CLI: o4-mini (90% on Patch, mapping to \$14,422). Codex CLI: o3-high, Codex CLI: o4-mini, and Claude Code are more capable at defense, achieving higher Patch scores of 90%, 90%, and 87.5%, compared to Exploit scores of 47.5%, 32.5%, and 57.5% respectively; while the custom agents are relatively balanced between offense and defense, achieving Exploit scores of 17.5-67.5% and Patch scores of 25-60%.

Transformer-based language models excel at both recall (retrieving memorized facts) and reasoning (performing multi-step inference), but whether these abilities rely on distinct internal mechanisms remains unclear. Distinguishing recall from reasoning is crucial for predicting model generalization, designing targeted evaluations, and building safer interventions that affect one ability without disrupting the other.We approach this question through mechanistic interpretability, using controlled datasets of synthetic linguistic puzzles to probe transformer models at the layer, head, and neuron level. Our pipeline combines activation patching and structured ablations to causally measure component contributions to each task type. Across two model families (Qwen and LLaMA), we find that interventions on distinct layers and attention heads lead to selective impairments: disabling identified "recall circuits" reduces fact-retrieval accuracy by up to 15\% while leaving reasoning intact, whereas disabling "reasoning circuits" reduces multi-step inference by a comparable margin. At the neuron level, we observe task-specific firing patterns, though these effects are less robust, consistent with neuronal polysemanticity.Our results provide the first causal evidence that recall and reasoning rely on separable but interacting circuits in transformer models. These findings advance mechanistic interpretability by linking circuit-level structure to functional specialization and demonstrate how controlled datasets and causal interventions can yield mechanistic insights into model cognition, informing safer deployment of large language models.

Modern LLM deployments confront a widening cost-performance spectrum: premium models deliver strong reasoning but are expensive, while lightweight models are economical yet brittle on complex tasks. Static escalation rules and keyword heuristics under-utilize this spectrum and fail to adapt across task types. We present xRouter, a tool-calling-based routing system in which a learned router can either answer directly or invoke one or more external models. The router is trained end-to-end with reinforcement learning using an explicit, cost-aware reward that encodes cost-performance trade-offs, eliminating the need for hand-engineered routing rules. Our implementation encompasses the full reinforcement learning framework, including reward and cost accounting, as well as the deployment and evaluation pipelines. Across diverse benchmarks, xRouter achieves strong cost-performance trade-offs (e.g., substantial cost reductions at comparable task completion rates), and provides empirical insights into what reliably helps learned routing and what does not, ranging from model trainability to the difficulty of eliciting sophisticated orchestration behaviors in small open models. We hope these findings and our open implementation will serve as a practical substrate for advancing learned, cost-aware LLM orchestration.

Activation-based steering can personalize large language models at inference time, but its effects in educational settings remain unclear. We study persona vectors for seven character traits in short-answer generation and automated scoring on the ASAP-SAS benchmark across three models spanning two architectures. Persona steering lowers answer quality overall, with much larger effects on open-ended English Language Arts (ELA) prompts than on factual science prompts; interpretive and argumentative tasks are up to 11x more sensitive. On the scoring side, we observe predictable valence-aligned calibration shifts: evil and impolite scorers grade more harshly, while good and optimistic scorers grade more leniently. ELA tasks are 2.5-3x more susceptible to scorer personalization than science tasks, and the Mixture-of-Experts model shows roughly 6x larger calibration shifts than the dense models. To our knowledge, this is the first study to systematically examine the effects of activation-steered persona traits in educational generation and scoring, and the results highlight the need for task-aware and architecture-aware calibration when deploying steered models in educational settings.

Instruction-guided image editing is becoming a general interface for visual work, yet existing benchmarks still focus largely on narrow appearance edits and do not fully capture the diversity of real-image tasks in professional workflows. Here, we define instructional computer vision problem solving as a broader formulation of image editing: given a real input image and a natural-language instruction, a system must produce an edited output that realizes the requested transformation while satisfying explicit preservation, geometric, physical, and usability constraints. We introduce CV-Arena, an open benchmark designed to evaluate this capability at professional scales. CV-Arena contains 12K high-resolution real-image instruction pairs spanning 16 instruction-based visual task types, constructed using CogRetriever, a dual-track retrieval-and-curation pipeline that combines targeted web search, agentic query refinement, verification, and traceability. To evaluate models at scale while preserving human fidelity, we propose Active Elo, a human-AI collaborative preference protocol that leverages CV-Judge, a logic-gated, multi-dimensional VLM evaluator, to reject clear failures and resolve high-confidence comparisons; and to route close, high-quality comparisons to expert raters. Mixed human and AI supervision is then aggregated through reliability-weighted Elo updates. Our comprehensive evaluation of 21 systems, including proprietary, open-source, and agentic models, on CV-Arena reveals persistent gaps in instruction adherence, physical reasoning, structural control, and fine-grained detail preservation. We further develop CV-Agent, a lightweight agentic model that combines planning, editing, and verification, and demonstrate that closed-loop reasoning is a promising direction for professional-grade instruction-following visual editing.

As large language models evolve from question-answering systems into general-purpose agents, evaluation must move beyond static answer correctness to assess multimodal perception, multi-step execution, tool use, and artifact delivery. However, existing benchmarks are often tied to specific task types, execution environments, or scoring protocols, limiting their comparability, interpretability, and reliability for deployment decisions. We introduce DAREBench (Deployment-Aware and Reliable Evaluation of Models as Agents), a benchmark designed to capture workload variation and support reliable agent evaluation. Built on a shared OpenClaw execution environment, DAREBench organizes 233 tasks selected and adapted from 22 source benchmarks into a 2times3 workload matrix defined by input modality and execution form, and evaluates them under a unified contract-based protocol with evidence-based score auditing. We evaluate 23 commercial API models and 12 locally deployed open-weight models over 7,587 model--task runs, reporting accuracy and token consumption alongside reference costs for API models. Results show that no single model dominates all workload groups, text and multimodal tasks exhibit distinct accuracy--cost trade-offs, and local open-weight models are competitive in several groups but still trail frontier commercial models overall. These findings suggest that agent deployment and model selection should consider workload profiles, deployment mode, and accuracy--cost trade-offs rather than rely on a single aggregate score.

Prefill-only decision models inspired by the Jev model score every candidate in a menu during a single forward pass and never decode, which makes one call one to two orders of magnitude cheaper than a same-scale generative language model. We show that this read-out structure comes with a testable property. When an intervention changes only the candidate menu and leaves the input text fixed, the post-intervention accuracy is already determined by the cached first-pass distribution. The estimator restricts the pass-1 probabilities to the menu, renormalizes, and reads off the argmax; it uses no labels and no second forward pass. Across seven model families, ten datasets and two task types, menu-only interventions are predicted to within 4.2 points, and for one family the prediction is exact. A probability-level variant of the same estimator errs by 21.0 points, so the property lives in the ranking rather than in the probabilities and is not recovered by calibration. Same-scale generative language models do not share the property. On those models the same estimator errs by 1.6 to 15.8 points and degrades as the model grows. The property turns inference-time compute into a decision that can be made before deployment. Uniform extra passes buy calibration but almost no accuracy; at matched cost a confidence cascade outperforms every scheme that re-asks the same model, and curating the menu beats enlarging the model, with a 0.8B model on a curated 5-candidate menu reaching 95.4% on CLINC150 against 80.0% for a 4B model on the full 150-label menu.Code and data are available at https://github.com/rlisml/jev-cascade.

Continual learning has gained increasing importance as it facilitates the acquisition and refinement of scalable knowledge and skills in language models. However, existing methods typically encounter strict limitations and challenges in real-world scenarios, such as reliance on experience replay, optimization constraints, and inference task-ID. In this study, we introduce the Scalable Language Model (SLM) to overcome these limitations within a more challenging and generalized setting, representing a significant advancement toward practical applications for continual learning. Specifically, we propose the Joint Adaptive Re-Parameterization (JARe), integrated with Dynamic Task-related Knowledge Retrieval (DTKR), to enable adaptive adjustment of language models based on specific downstream tasks. This approach leverages the task distribution within the vector space, aiming to achieve a smooth and effortless continual learning process. Our method demonstrates state-of-the-art performance on diverse backbones and benchmarks, achieving effective continual learning in both full-set and few-shot scenarios with minimal forgetting. Moreover, while prior research primarily focused on a single task type such as classification, our study goes beyond, with the large language model, i.e., LLaMA-2, to explore the effects across diverse domains and task types, such that a single language model can be decently scaled to broader applications.

We present BenchRL-QAS, a unified benchmarking framework for reinforcement learning (RL) in quantum architecture search (QAS) across a spectrum of variational quantum algorithm tasks on 2- to 8-qubit systems. Our study systematically evaluates 9 different RL agents, including both value-based and policy-gradient methods, on quantum problems such as variational eigensolver, quantum state diagonalization, variational quantum classification (VQC), and state preparation, under both noiseless and noisy execution settings. To ensure fair comparison, we propose a weighted ranking metric that integrates accuracy, circuit depth, gate count, and training time. Results demonstrate that no single RL method dominates universally, the performance dependents on task type, qubit count, and noise conditions providing strong evidence of no free lunch principle in RL-QAS. As a byproduct we observe that a carefully chosen RL algorithm in RL-based VQC outperforms baseline VQCs. BenchRL-QAS establishes the most extensive benchmark for RL-based QAS to date, codes and experimental made publicly available for reproducibility and future advances.

Large language models (LLMs) demonstrate remarkable capabilities, but their broad deployment is limited by significant computational resource demands, particularly energy consumption during inference. Static, one-model-fits-all inference strategies are often inefficient, as they do not exploit the diverse range of available models or adapt to varying query requirements. This paper presents GreenServ, a dynamic, context-aware routing framework that optimizes the trade-off between inference accuracy and energy efficiency. GreenServ extracts lightweight contextual features from each query, including task type, semantic cluster, and text complexity, and routes queries to the most suitable model from a heterogeneous pool, based on observed accuracy and energy usage. We employ a multi-armed bandit approach to learn adaptive routing policies online. This approach operates under partial feedback, eliminates the need for extensive offline calibration, and streamlines the integration of new models into the inference pipeline. We evaluated GreenServ across five benchmark tasks and a pool of 16 contemporary open-access LLMs. Experimental results show that GreenServ consistently outperforms static (single-model) and random baselines. In particular, compared to random routing, GreenServ achieved a 22% increase in accuracy while reducing cumulative energy consumption by 31%. Finally, we evaluated GreenServ with RouterBench, achieving an average accuracy of 71.7% with a peak accuracy of 75.7%. All artifacts are open-source and available here: https://github.com/TZData1/llm-inference-router{GitHub}

Persona prompting can steer LLM generation towards a domain-specific tone and pattern. This behavior enables use cases in multi-agent systems where diverse interactions are crucial and human-centered tasks require high-level human alignment. Prior works provide mixed opinions on their utility: some report performance gains when using expert personas for certain domains and their contribution to data diversity in synthetic data creation, while others find near-zero or negative impact on general utility. To fully leverage the benefits of the LLM persona and avoid its harmfulness, a more comprehensive investigation of the mechanism is crucial. In this work, we study how model optimization, task type, prompt length, and placement can impact expert persona effectiveness across instruction-tuned and reasoning LLMs, and provide insight into conditions under which expert personas fail and succeed. Based on our findings, we developed a pipeline to fully leverage the benefits of an expert persona, named PRISM (Persona Routing via Intent-based Self-Modeling), which self-distills an intent-conditioned expert persona into a gated LoRA adapter through a bootstrapping process that requires no external data, models, or knowledge. PRISM enhances human preference and safety alignment on generative tasks while maintaining accuracy on discriminative tasks across all models, with minimal memory and computing overhead.

Recent advances in singing voice synthesis (SVS) have attracted substantial attention from both academia and industry. With the advent of large language models and novel generative paradigms, producing controllable, high-fidelity singing voices has become an attainable goal. Yet the field still lacks a comprehensive survey that systematically analyzes deep-learning-based singing voice synthesis systems and their enabling technologies. To address the aforementioned issue, this survey first categorizes existing systems by task type and then organizes current architectures into two major paradigms: cascaded and end-to-end approaches. Moreover, we provide an in-depth analysis of core technologies, covering singing modeling and control techniques. Finally, we review relevant datasets, annotation tools, and evaluation benchmarks that support training and assessment. In appendix, we introduce training strategies and further discussion of SVS. This survey provides an up-to-date review of the literature on SVS models, which would be a useful reference for both researchers and engineers. Related materials are available at https://github.com/David-Pigeon/SyntheticSingers.

We introduce an axiomatic evaluation framework for latent thought representations in LLMs, comprising metrics that are independent of downstream benchmark scores and reveal representational failures that benchmark accuracy masks. Existing evaluations conflate representation quality with model capacity. Therefore, failures cannot be attributed to the representation rather than to the model that processes it. We formalize four functional axioms (Causality, Minimality, Separability, and Stability) and define a quantitative measure for each, computed directly on the representation independently of downstream accuracy. We audit open-weight LLMs across 23 reasoning tasks (e.g., Spatial Reasoning, Factual QA). We find that no candidate satisfies all four axioms simultaneously, that the representations distinguish task type reliably but cannot distinguish between two questions within the same task, and that the representations encode little information beyond what is already present in the input embedding. The failure is consistent across dense, reasoning-distilled, and RL-trained model families, indicating that the gap is structural rather than a property of model size or training procedure.

Large Language Models (LLMs) are popular for their impressive abilities, but the need for model-specific fine-tuning or task-specific prompt engineering can hinder their generalization. We propose UPRISE (Universal Prompt Retrieval for Improving zero-Shot Evaluation), which tunes a lightweight and versatile retriever that automatically retrieves prompts for a given zero-shot task input. Specifically, we demonstrate universality in a cross-task and cross-model scenario: the retriever is tuned on a diverse set of tasks, but tested on unseen task types; we use a small frozen LLM, GPT-Neo-2.7B, for tuning the retriever, but test the retriever on different LLMs of much larger scales, such as BLOOM-7.1B, OPT-66B and GPT3-175B. Additionally, we show that UPRISE mitigates the hallucination problem in our experiments with ChatGPT, suggesting its potential to improve even the strongest LLMs. Our model and code are available at https://github.com/microsoft/LMOps.

Task-specific deep learning models in histopathology offer promising opportunities for improving diagnosis, clinical research, and precision medicine. However, development of such models is often limited by availability of high-quality data. Foundation models in histopathology that learn general representations across a wide range of tissue types, diagnoses, and magnifications offer the potential to reduce the data, compute, and technical expertise necessary to develop task-specific deep learning models with the required level of model performance. In this work, we describe the development and evaluation of foundation models for histopathology via self-supervised learning (SSL). We first establish a diverse set of benchmark tasks involving 17 unique tissue types and 12 unique cancer types and spanning different optimal magnifications and task types. Next, we use this benchmark to explore and evaluate histopathology-specific SSL methods followed by further evaluation on held out patch-level and weakly supervised tasks. We found that standard SSL methods thoughtfully applied to histopathology images are performant across our benchmark tasks and that domain-specific methodological improvements can further increase performance. Our findings reinforce the value of using domain-specific SSL methods in pathology, and establish a set of high quality foundation models to enable further research across diverse applications.

Understanding and reasoning over diagrams is a fundamental aspect of human intelligence. While Large Multimodal Models (LMMs) have demonstrated impressive capabilities across various tasks, existing benchmarks lack comprehensive evaluation of their diagram interpretation and reasoning abilities, particularly in coding contexts. We present HumanEval-V, a rigorous benchmark of human-annotated coding tasks that spans six task types and evaluates diverse visual reasoning capabilities. Each task features carefully crafted diagrams paired with function signatures and test cases, employing novel code generation tasks to thoroughly assess models' diagram comprehension. Through extensive experiments with 22 LMMs, we find that even top-performing models achieve modest success rates, with Claude 3.5 Sonnet reaching only 36.8% pass@1, highlighting substantial room for improvement. Our analysis reveals that current LMMs struggle with spatial transformations, topological relationships, and dynamic patterns that humans find intuitive. These findings provide valuable insights for advancing LMMs' visual reasoning abilities. We have open-sourced our code and benchmark at https://github.com/HumanEval-V/HumanEval-V-Benchmark.

Artificial Intelligence (AI), with ChatGPT as a prominent example, has recently taken center stage in various domains including higher education, particularly in Computer Science and Engineering (CSE). The AI revolution brings both convenience and controversy, offering substantial benefits while lacking formal guidance on their application. The primary objective of this work is to comprehensively analyze the pedagogical potential of ChatGPT in CSE education, understanding its strengths and limitations from the perspectives of educators and learners. We employ a systematic approach, creating a diverse range of educational practice problems within CSE field, focusing on various subjects such as data science, programming, AI, machine learning, networks, and more. According to our examinations, certain question types, like conceptual knowledge queries, typically do not pose significant challenges to ChatGPT, and thus, are excluded from our analysis. Alternatively, we focus our efforts on developing more in-depth and personalized questions and project-based tasks. These questions are presented to ChatGPT, followed by interactions to assess its effectiveness in delivering complete and meaningful responses. To this end, we propose a comprehensive five-factor reliability analysis framework to evaluate the responses. This assessment aims to identify when ChatGPT excels and when it faces challenges. Our study concludes with a correlation analysis, delving into the relationships among subjects, task types, and limiting factors. This analysis offers valuable insights to enhance ChatGPT's utility in CSE education, providing guidance to educators and students regarding its reliability and efficacy.