Hugging Face's logo

MiniMaxAI
/
MiniMax-M3.1-Flash

Image-Text-to-Text
Transformers
Safetensors
minimax_m3_vl
multimodal
Mixture of Experts
agent
coding
video
conversational
custom_code
Eval Results
8-bit precision
mxfp8
Model card Files
xet
 Community
6
MiniMax-M3.1-Flash
File size: 23,407 Bytes
54e4e69
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
---
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)。