| --- |
| license: apache-2.0 |
| base_model: Qwen/Qwen3.8-Flash-Next |
| tags: |
| - qwen |
| - mach-2 |
| - additive |
| - mixture-of-experts |
| --- |
| |
|  |
|
|
| # Mach-2 Additive Medium |
|
|
| Mach-2 Additive Medium is Qwen3.8-Flash-Next at **1.70 bits per weight**: 26.7 GB for the 125.7B text parameters, plus |
| the model's n-gram embedding table at bf16 (102.4 GB). |
|
|
| ## Benchmarks |
|
|
|  |
|
|
| Retention is Mach-2 Additive Medium's score divided by that of Qwen3.8-Flash-Next BF16, with both run on the same |
| harness, settings and tasks. |
|
|
| | Benchmark | Tasks | Mach-2 Additive Medium | BF16 | Retention | |
| | --- | ---: | ---: | ---: | ---: | |
| | Humanity's Last Exam (text) | 2,158 | 33.87 | 39.85 | 85.0% | |
| | GPQA Diamond | 198 x 5 | 90.51 | 92.22 | 98.1% | |
| | AIME 2026 | 30 x 8 | 96.25 | 95.83 | 100.4% | |
| | Terminal-Bench 2.1 | 89 | 78.65 | 82.02 | 95.9% | |
| | AutomationBench 1.0.6 | 600 | 71.67 | 70.29 | 102.0% | |
| | NL2Repo | 98 | 56.12 | 60.06 | 93.4% | |
| | DeepSWE v1.1 | 113 | 48.67 | 49.56 | 98.2% | |
| | τ³-Banking | 97 x 5 | 45.10 | 46.41 | 97.2% | |
|
|
| - **Humanity's Last Exam**: up to 163,840 output tokens, graded by Artificial Analysis's judge. |
| - **GPQA Diamond**: mean accuracy over 5 samples per question. Answers that reached 81,920 tokens were re-asked at |
| 163,840. |
| - **AIME 2026**: mean accuracy over 8 samples per problem, up to 81,920 output tokens. |
| - **Terminal-Bench 2.1**: Terminus 2, one attempt per task, 4 hours per task. |
| - **AutomationBench**: up to 50 steps per task, scored by Artificial Analysis's rule. |
| - **NL2Repo**: OpenHands without internet access; the 102 tasks BF16 finished, less 4 whose BF16 test run hit the |
| 1-hour scoring limit. Test runs cut at that limit are left out rather than scored 0. Mach-2 Additive Medium's score |
| averages two runs. |
| - **DeepSWE**: Claude Code as the agent, all 113 tasks, up to 10.5 hours per task. A task counts as solved only if the |
| agent solved it within 700 steps. |
| - **τ³-Banking**: tau2-bench 1.0.1, 5 trials per task, GPT-5.4 mini as the simulated user and judge, as Artificial |
| Analysis runs it. Pass rate over the 459 conversations both models completed; 26 that outgrew the context window or |
| lost the simulated user to an API error are left out. |
|
|
| Empty answers and answers still unfinished at the token limit score 0. Agents and limits differ from those behind |
| Qwen's model card, so compare these columns with each other rather than with scores published elsewhere. |
|
|
| ## Use |
|
|
| Run Mach-2 Additive Medium from its compressed weights with |
| [Mach-2 Additive Medium GGUF](https://huggingface.co/SyzygyResearch/Mach-2-Additive-Medium-GGUF) and the |
| [Mach-1 fork of llama.cpp](https://github.com/SyzygyResearch/llama.cpp-mach1). The 26.7 GB of weights fit on one GPU |
| with 32 GB or more. |
|
|
| Sample at temperature 1.0, top_p 0.95, top_k 20 (`generation_config.json`). Without these settings the model can loop. |
|
|
| This repository holds the packed weights; `decode.py` documents the packed format. `draft/mtp_q4g64.safetensors` is the |
| base model's MTP layer at 4 bits (group size 64) for speculative decoding with MLX. |
|
|
