| --- |
| pipeline_tag: image-text-to-text |
| license: other |
| license_name: minimax-community |
| license_link: LICENSE |
| library_name: transformers |
| tags: |
| - multimodal |
| - moe |
| - agent |
| - coding |
| - video |
| --- |
| |
|
|
| <p align="right"><a href="README.md">English</a></p> |
|
|
| # MiniMax-M3.1-Flash |
|
|
| <div align="center"> |
| <img width="60%" src="figures/logo.svg" alt="MiniMax"> |
| </div> |
| <hr> |
|
|
| <p align="center"> |
| <a href="https://agent.minimax.io/" target="_blank"><img src="https://img.shields.io/badge/MiniMax%20Agent-FF6C37?logo=minimax&logoColor=white" alt="MiniMax Agent"></a> |
| <a href="https://platform.minimax.io/docs/guides/text-generation" target="_blank"><img src="https://img.shields.io/badge/API-FF6C37?logo=minimax&logoColor=white" alt="API"></a> |
| <a href="https://www.minimax.io" target="_blank"><img src="https://img.shields.io/badge/MiniMax%20Website-FF6C37?logo=minimax&logoColor=white" alt="MiniMax Website"></a> |
| <br> |
| <a href="https://modelscope.cn/organization/minimax" target="_blank" rel="noopener noreferrer"><img alt="ModelScope MiniMax AI" src="https://img.shields.io/badge/ModelScope-MiniMax%20AI-white?labelColor=%23EF3D5D"></a> |
| <a href="https://platform.minimaxi.com/docs/faq/contact-us" target="_blank"><img src="https://img.shields.io/badge/WeChat-07C160?logo=wechat&logoColor=white" alt="WeChat"></a> |
| <a href="https://discord.com/invite/DPC4AHFCBw" target="_blank"><img src="https://img.shields.io/badge/Discord-5865F2?logo=discord&logoColor=white" alt="Discord"></a> |
| <a href="https://huggingface.co/MiniMaxAI" target="_blank"><img src="https://img.shields.io/badge/Hugging%20Face-FFD21E?logo=huggingface&logoColor=black" alt="Hugging Face"></a> |
| <a href="https://github.com/MiniMax-AI" target="_blank"><img src="https://img.shields.io/badge/GitHub-181717?logo=github&logoColor=white" alt="GitHub"></a> |
| <a href="https://arxiv.org/abs/2606.13392" target="_blank"><img src="https://img.shields.io/badge/arXiv-2606.13392-B31B1B?logo=arxiv&logoColor=white" alt="arXiv Paper"></a> |
| </p> |
|
|
| **MiniMax-M3.1-Flash** 是一款原生多模态混合专家（MoE）模型，面向编程、智能体工作流、计算机使用和网络安全等场景。它支持文本、图像和视频输入，上下文窗口最高可达 **100 万 tokens**。 |
|
|
| 在 MiniMax-M3 的基础上，M3.1-Flash 通过全模型稀疏注意力、量化感知训练和 DSpark 投机解码，进一步提升了任务表现与推理效率。 |
|
|
| **亮点：** |
| - **智能体能力：** 在智能体任务、编程和网络安全方面实现代际跃升——BankerToolBench 76.2、DeepSWE v1.1 73.7%、OSWorld V2 61.7、Agents' Last Exam 29.6、CyberGym 82.2。 |
| - **高效设计：** 全模型稀疏注意力、量化感知训练和 DSpark 投机解码共同提升推理效率。 |
|
|
| ## 模型概览 |
|
|
| - **架构：** 原生多模态混合专家（MoE） |
| - **参数：** 总参数约 428B；每 token 激活约 23B |
| - **输入 / 输出：** 文本、图像和视频输入；文本输出 |
| - **上下文窗口：** 最高 1,000,000 tokens |
| - **注意力：** 全模型采用 [MiniMax Sparse Attention](https://arxiv.org/abs/2606.13392)（MSA） |
| - **投机解码：** 包含 DSpark draft head |
|
|
| ## 基准测试 |
|
|
| <div style="overflow-x:auto;margin:16px 0"> |
| <table style="width:100%;border-collapse:collapse;font-size:14px"> |
| <thead> |
| <tr> |
| <th style="padding:10px 12px;text-align:left;vertical-align:middle;border-bottom:1px solid #e5e7eb;font-weight:600;border-bottom:2px solid #f25f5c;background:#f7f7f8;">基准测试</th> |
| <th style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;font-weight:600;border-bottom:2px solid #f25f5c;background:rgba(242,95,92,0.08);">MiniMax M3.1 Flash</th> |
| <th style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;font-weight:600;border-bottom:2px solid #f25f5c;background:#f7f7f8;">DeepSeek V4.1 Flash-1017</th> |
| <th style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;font-weight:600;border-bottom:2px solid #f25f5c;background:#f7f7f8;">DeepSeek V4.1 Flash</th> |
| <th style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;font-weight:600;border-bottom:2px solid #f25f5c;background:#f7f7f8;">Kimi K3</th> |
| <th style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;font-weight:600;border-bottom:2px solid #f25f5c;background:#f7f7f8;">GLM-5.3</th> |
| <th style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;font-weight:600;border-bottom:2px solid #f25f5c;background:#f7f7f8;">GPT-5.6 Sol</th> |
| <th style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;font-weight:600;border-bottom:2px solid #f25f5c;background:#f7f7f8;">Opus 5</th> |
| </tr> |
| </thead> |
| <tbody> |
| <tr> |
| <td style="padding:10px 12px;text-align:left;vertical-align:middle;border-bottom:1px solid #e5e7eb;"><div>Automation-Bench v1.0.6</div></td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;background:rgba(242,95,92,0.08);"><strong>62.8</strong></td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">60.5</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">54.8</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">46.7</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">48.8</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">45.8</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">50.3</td> |
| </tr> |
| <tr> |
| <td style="padding:10px 12px;text-align:left;vertical-align:middle;border-bottom:1px solid #e5e7eb;">BankerToolBench</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;background:rgba(242,95,92,0.08);"><strong>76.2</strong></td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">—</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">73.7</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">69.1</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">74.5</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">74.6</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;"><strong>76.2</strong></td> |
| </tr> |
| <tr> |
| <td style="padding:10px 12px;text-align:left;vertical-align:middle;border-bottom:1px solid #e5e7eb;">GDP.pdf AA</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;background:rgba(242,95,92,0.08);">20.8</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">—</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">12.8</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">22.0</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">11.2</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;"><strong>27.2</strong></td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">21.6</td> |
| </tr> |
| <tr> |
| <td style="padding:10px 12px;text-align:left;vertical-align:middle;border-bottom:1px solid #e5e7eb;">Terminal Bench 4.0</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;background:rgba(242,95,92,0.08);">38.1</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">42.3</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">31.2</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">12.6</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">37.9</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">39.9</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;"><strong>51.8</strong></td> |
| </tr> |
| <tr> |
| <td style="padding:10px 12px;text-align:left;vertical-align:middle;border-bottom:1px solid #e5e7eb;">LOCA-Bench 256K</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;background:rgba(242,95,92,0.08);"><strong>60.3</strong></td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">—</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">—</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">49.7</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">50.3</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">60.0</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">53.7</td> |
| </tr> |
| <tr> |
| <td style="padding:10px 12px;text-align:left;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;"><div>Agents’ Last Exam</div></td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;background:rgba(242,95,92,0.08);">29.6</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;">—</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;">—</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;">28.3</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;">—</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;">30.6</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;"><strong>32.2</strong></td> |
| </tr> |
| <tr> |
| <td style="padding:10px 12px;text-align:left;vertical-align:middle;border-bottom:1px solid #e5e7eb;">OSWorld 2.0</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;background:rgba(242,95,92,0.08);">61.7</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">—</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">54.3</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">58.3</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">—</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">65.7</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;"><strong>70.6</strong></td> |
| </tr> |
| <tr> |
| <td style="padding:10px 12px;text-align:left;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;"><div>PostTrain Bench v1.1</div></td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;background:rgba(242,95,92,0.08);">39.3</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;">—</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;">30.5</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;">36.6</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;">37.5</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;">36.2</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;"><strong>40.4</strong></td> |
| </tr> |
| <tr> |
| <td style="padding:10px 12px;text-align:left;vertical-align:middle;border-bottom:1px solid #e5e7eb;">DeepSWE v1.1</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;background:rgba(242,95,92,0.08);">73.7</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;"><strong>77.0</strong></td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">74.2</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">67.5</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">66.9</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">73.0</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">74.0</td> |
| </tr> |
| <tr> |
| <td style="padding:10px 12px;text-align:left;vertical-align:middle;border-bottom:1px solid #e5e7eb;">NL2Repo-Bench</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;background:rgba(242,95,92,0.08);">61.2</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">68.2</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">65.4</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">58.0</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">58.0</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">56.8</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;"><strong>75.3</strong></td> |
| </tr> |
| <tr> |
| <td style="padding:10px 12px;text-align:left;vertical-align:middle;border-bottom:1px solid #e5e7eb;">SWE Refactor Bench</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;background:rgba(242,95,92,0.08);">23.0</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">—</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">11.8</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">19.5</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">10.4</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">28.5</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;"><strong>31.0</strong></td> |
| </tr> |
| <tr> |
| <td style="padding:10px 12px;text-align:left;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;"><div>CyberGym</div></td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;background:rgba(242,95,92,0.08);">82.2</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;">—</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;">—</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;">80.0</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;"><strong>84.5</strong></td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;">83.6</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;">—</td> |
| </tr> |
| <tr> |
| <td style="padding:10px 12px;text-align:left;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;"><div>BabyVision (with tools)</div></td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;background:rgba(242,95,92,0.08);">91.0</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;">92.5</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;">89.6</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;">85.7</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;">—</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;">88.9</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;border-top:2px solid #f1d6d3;"><strong>94.1</strong></td> |
| </tr> |
| <tr> |
| <td style="padding:10px 12px;text-align:left;vertical-align:middle;border-bottom:1px solid #e5e7eb;">HLE (without tools)</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;background:rgba(242,95,92,0.08);">43.4</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">42.6</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">39.2</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">43.5</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">42.3</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">49.5</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;"><strong>55.6</strong></td> |
| </tr> |
| <tr> |
| <td style="padding:10px 12px;text-align:left;vertical-align:middle;border-bottom:1px solid #e5e7eb;">CharXiv RQ</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;background:rgba(242,95,92,0.08);">82.5</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">—</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">69.0</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">84.8</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">—</td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;"><strong>85.2</strong></td> |
| <td style="padding:10px 12px;text-align:center;vertical-align:middle;border-bottom:1px solid #e5e7eb;">83.7</td> |
| </tr> |
| </tbody> |
| </table> |
| </div> |
|
|
| <details> |
| <summary><strong>评测说明</strong></summary> |
|
|
| - **Automation-Bench v1.0.6：** 600 个公开任务。 |
| - **BankerToolBench：** OpenCode 1.17.18；MiniMax-M2.7 作为 judge。 |
| - **GDP.pdf AA：** 直接取自 Artificial Analysis。 |
| - **Terminal Bench 4.0：** Codex 0.149.1。 |
| - **LOCA-Bench 256K：** 官方实现。 |
| - **Agents’ Last Exam：** 采用官方 Overall 设置；仅收录官方 ALE 榜单的完整测试集结果。空白表示没有官方结果（仅 CLI 结果不计入），并非零分。 |
| - **OSWorld 2.0：** 采用 0808 Full 版本；各模型取其发布时报告的最高分，DeepSeek V4.1 除外（内部复现）。 |
| - **PostTrain Bench v1.1：** Claude Code 2.1.251，512K 上下文；其余设置遵循官方评测。 |
| - **DeepSWE v1.1：** mini-SWE-agent 2.4.5，512K 上下文。 |
| - **NL2Repo-Bench：** Claude Code 2.1.233，512K 上下文；reward hacking 计 0 分；修正了两个规格与测试不一致的任务（分数基于修正后的测试集）。 |
| - **SWE Refactor Bench：** Claude Code 2.1.220，200K 上下文；其余设置遵循官方评测。 |
| - **CyberGym：** Claude Code 2.1.263，1M 上下文。 |
| - **BabyVision（带工具）：** Claude Code 2.1.285；图像最短边 1,500 px。 |
| - **HLE（不带工具）：** 纯文本子集（2,158 题）。 |
| - **CharXiv RQ：** GPT-5.5 作为 judge。 |
|
|
| </details> |
|
|
| 以上结果基于官方权重；第三方量化版本可能有所不同。 |
|
|
| ## M3.1-Flash 的新特性 |
|
|
| M3.1-Flash 延续 M3 的核心混合专家架构，主要有三项变化： |
|
|
| - **全模型稀疏注意力：** 前三个全注意力层替换为稀疏注意力。 |
| - **量化感知训练：** 路由 MoE 专家与注意力路径在训练阶段完成量化，降低显存占用并提升推理效率。 |
| - **DSpark 投机解码：** DSpark 取代此前的类 EAGLE MTP 方案。本次发布包含 draft head，可与兼容的 serving 实现配合使用。 |
|
|
| ## 如何使用 |
|
|
| - [MiniMax Agent](https://agent.minimax.io/) |
| - [MiniMax API](https://platform.minimax.io/) |
|
|
| ### 本地部署 |
|
|
| 下载模型： |
|
|
| ```bash |
| hf download MiniMaxAI/MiniMax-M3.1-Flash --local-dir MiniMax-M3.1-Flash |
| ``` |
|
|
| 推荐使用以下推理框架部署模型： |
|
|
| - [SGLang](https://docs.sglang.io/) — 参见 [SGLang cookbook](https://docs.sglang.io/cookbook/autoregressive/MiniMax/MiniMax-M3.1)。 |
| - [vLLM](https://github.com/vllm-project/vllm) — 参见 [vLLM recipe](https://recipes.vllm.ai/MiniMaxAI/MiniMax-M3.1-Flash)。 |
|
|
| 更多集成正在推进中：Transformers、TokenSpeed、KTransformers、unsloth 和 ATOM。 |
|
|
| ## 推理设置 |
|
|
| 推荐采样设置：`temperature=1.0` 和 `top_p=0.95`。 |
|
|
| 推理时，在 chat template 中设置 `reasoning_effort`：`low`、`medium`、`high`、`xhigh` 或 `max`。 |
|
|
| 模型支持最高 100 万 tokens 的上下文窗口。可用上下文长度和吞吐量取决于 GPU 显存和 serving 配置。 |
|
|
| ## 联系我们 |
|
|
| 请联系 [model@minimax.io](mailto:model@minimax.io)。 |
|
|
