编排:Minimax-m3

这不是客观事实。 本页记录本站的梯队、星级、六维和情绪等主观判断，与模型事实页物理分离，仅编排员和管理员可修改。

<syntaxhighlight lang="yaml"> id: minimax-m3 tier: T1 stars: 5 composite: 89 stats:

 code: 80
 reasoning: 45
 context: 88
 speed: 65
 multimodal: 88
 value: 78

sentiment:

 positive: 40
 mixed: 25
 negative: 35

radar: - axis: 长程任务

 value: 85

- axis: 编程工程

 value: 80

- axis: 抽象推理

 value: 45

- axis: 上下文利用

 value: 88

- axis: 中文能力

 value: 88

- axis: 响应速度

 value: 65

- axis: 稳定性

 value: 45

- axis: 指令遵循

 value: 62

- axis: 易用性

 value: 72

- axis: 性价比

 value: 78

danmaku: - text: Another model that has vision

 platform: reddit
 main: true

- text: thinks for THOUSANDS of tokens

 platform: reddit
 main: true

- text: prefer it to opus in agentic workflows

 platform: reddit
 main: true

- text: M3 not far behind DS4

 platform: hn
 main: true

- text: 平替天花板，体感超 Sonnet 4.6

 platform: zhihu
 main: true

- text: 又背刺老用户，骂声一片

 platform: v2ex
 main: true

- text: token plan is very cheap

 platform: hn
 main: false

- text: It's open weights, not open source

 platform: hn
 main: false

- text: M2.7 出好代码，M3 老出语法错误

 platform: reddit
 main: false

- text: 实际成本 49 涨到 175，+257%

 platform: v2ex
 main: false

- text: 1M 上下文真正可用而非仅标注

 platform: zhihu
 main: false

- text: MSA 支持已合入 llama.cpp

 platform: hn
 main: false

- text: 2.7T 参数模型在路上

 platform: reddit
 main: false

- text: 长任务用 M3，小快活交给 M2.7

 platform: reddit
 main: false

- text: M3?!!! So gooooooood!!!

 platform: x
 main: false

</syntaxhighlight>