Microsoft should've released something like Qwen3.6-27B / Gemma-4-31B already. They released MAI models now

In three linesMicrosoft launches 7 MAI models: MAI-Thinking-1 (1T params, 256K context) matches leading models on software engineering benchmarks; MAI-Code-1-Flash (5B active params) integrated into GitHub Copilot and VS Code; MAI-Image-2.5 for text-to-image and editing; MAI-Transcribe-1.5 (SOTA, 5x faster); MAI-Voice-2 for speech generation. No open weights announced, proprietary licenses.

Summary generated by Claude — human-verified