---
license: mit
base_model: Qwen/Qwen3-8B
library_name: transformers
pipeline_tag: text-generation
tags:
- agent
- tool-use
- function-calling
- reflection
- reinforcement-learning
- arxiv:2610.17953
datasets:
- dongguanting/Agent-Reflex-SFT-54K
- dongguanting/Agent-Reflex-RL-2K
---

# Agent-Reflex-8B

[![Paper](https://img.shields.io/badge/Paper-arXiv-b5212f.svg?logo=arxiv)](https://arxiv.org/abs/2610.17953)
[![HF Paper](https://img.shields.io/badge/🤗%20Paper-2610.17953-ffc107.svg)](https://huggingface.co/papers/2610.17953)
[![GitHub](https://img.shields.io/badge/GitHub-Agent--Reflex-181717.svg?logo=github)](https://github.com/RUC-NLPIR/Agent-Reflex)
[![Collection](https://img.shields.io/badge/🤗%20Collection-Agent--Reflex-ffc107.svg)](https://huggingface.co/collections/dongguanting/agent-reflex)
[![SFT Data](https://img.shields.io/badge/Data-SFT--54K-ffc107.svg?logo=huggingface)](https://huggingface.co/datasets/dongguanting/Agent-Reflex-SFT-54K)
[![RL Data](https://img.shields.io/badge/Data-RL--2K-ffc107.svg?logo=huggingface)](https://huggingface.co/datasets/dongguanting/Agent-Reflex-RL-2K)

**Paper:** [arXiv:2610.17953](https://arxiv.org/abs/2610.17953) · [Hugging Face Papers](https://huggingface.co/papers/2610.17953)

Agent-Reflex-8B is Qwen3-8B trained with **Agent-Reflex**, a framework that elicits proactive reflection in tool-using agents: noticing noisy tool feedback and user corrections, and recovering from them.

Training: Qwen3-8B → **ReSFT** (54,663 reflective trajectories, 5,127 steps) → **ORPO** (noisy agentic RL with skill-guided on-policy distillation). Skills are used only during training; the model runs skill-free.

## Results

| Model | VitaBench | τ²-Bench | BFCL V4 | ACEBench | ToolSandbox | **Overall** | AgentNoiseBench |
| --- | :---: | :---: | :---: | :---: | :---: | :---: | :---: |
| Qwen3-8B | 11.4 | 26.2 | 40.4 | 54.0 | 69.1 | 40.2 | 12.3 |
| EnvScaler-8B | 15.8 | 37.9 | 47.6 | 72.5 | 80.5 | 50.9 | 18.2 |
| Agent-World-8B | 14.5 | 54.6 | 51.4 | 72.0 | 75.6 | 53.6 | – |
| **Agent-Reflex-8B** | **17.5** | **62.1** | **52.9** | **78.0** | **82.5** | **58.6** | **29.1** |

Each benchmark is scored with its official metric. Overall is the unweighted mean of the five long-horizon tool-use benchmarks (Table 1 of the paper); AgentNoiseBench is the mean over its τ²-Bench and VitaBench noise settings (Table 2).

## Usage

Serve with vLLM (native function calling, thinking on), as in our evaluation:

```bash
vllm serve dongguanting/Agent-Reflex-8B --served-model-name Qwen/Qwen3-8B \
  --tensor-parallel-size 8 --max-model-len 40960 \
  --enable-auto-tool-choice --tool-call-parser hermes --reasoning-parser deepseek_r1
```

The full evaluation suite (τ²-Bench, BFCL V4, ACEBench, VitaBench, AgentNoiseBench, ToolSandbox) is in the [GitHub repo](https://github.com/RUC-NLPIR/Agent-Reflex/tree/main/evaluation).

## Citation

```bibtex
@misc{dong2026agentreflex,
  title={Agent-Reflex: Teaching Language Agents to Act Reflectively},
  author={Guanting Dong and Xinyu Yang and Jiajie Jin and Yuyang Hu and Xiaoshuai Song and Yifei Chen and Xiaoxi Li and Jiejun Tan and Yutao Zhu and Tat-Seng Chua and Ji-Rong Wen and Zhicheng Dou},
  year={2026},
  eprint={2610.17953},
  archivePrefix={arXiv},
  primaryClass={cs.CL},
  url={https://arxiv.org/abs/2610.17953},
}
```