Unverified50% confidenceBenchmarkExact time
Optimizing MiniMax M3 inference serving on the MI355X ultimately achieved a 3–4x improvement in serving throughput.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/15/2026
First Seen
Valid until: 12/14/2026
Sources
Related Entities
Related Claims
UnverifiedThe ratio of activated parameters to total parameters for MiniMax M3, 23B/428B, is approximately 5.4%.66% 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65% 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65% similarUnverifiedAmong open-source models, the best-performing MiniMax M3 and Kimi achieved a completion rate of only 4.6% on OSWorld 2.0.63% similarUnverifiedIn the complex codebase feature expansion task, when Composer 2.5 had already completed the task, MiniMax M3 was still analyzing the codebase, but the final output contained more comments and more comprehensive test coverage.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/921464API
curl https://kongchang.com/api/v1/knowledge/claims/921464MCP
get_claim(id=921464)