Unverified50% confidenceFactExact time
MiniMax M3在复杂代码库功能扩展任务中，当Composer 2.5已完成任务时，MiniMax M3仍在分析代码库，但最终产出包含更多注释和更完善的测试覆盖
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
Valid until: 9/30/2026
Sources
MiniMax M3编程实测：价格仅为Opus的1/765，代码质量却更高？
bilibili认真的笨笨6/22/2026
Related Claims
UnverifiedMiniMax M3 received an average score of 58.3 across 7 high-difficulty coding tasks in a systematic evaluation74% similarUnverifiedMiniMax M2.7采用230亿参数的稀疏MoE（混合专家）架构，每个token仅激活100亿参数69% similarUnverified在MiniMax的MMClaw评估中，M2.7在开放式工具调用场景下接近Sonnet 4.6的水平67% similarUnverifiedMiniMax M2.1在SWE-Bench基准测试上的表现持平甚至超过Claude Sonnet 4.567% similarUnverified演示中使用的是MiniMax M3模型，而非市面上性能最强的模型，选型标准为够用、成本可靠66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/41665API
curl https://kongchang.com/api/v1/knowledge/claims/41665MCP
get_claim(id=41665)