2026-06-02 Chat Group Daily
2026-06-02 群聊日报
The chat group daily says Microsoft released MAI-Thinking-1 with 35B active parameters and about 1T MoE, matching Opus 4.6 on SWE-Bench Pro and scoring 97% on AIME 2025.
Why it matters: Microsoft's in-house reasoning model MAI-Thinking-1 matches Claude Opus 4.6 on SWE-Bench Pro coding with 35B active parameters, hits 97% on AIME 2025 math, and was trained without distilling from third-party models.