Microsoft unveils seven new AI models including first reasoning model

Microsoft opened its Build developer keynote by announcing seven new AI models, highlighted by MAI-Thinking-1, the lab's first reasoning model. CEO Mustafa Suleyman framed the rollout with the lab's "humanist superintelligence" language as he introduced the lineup.

MAI-Thinking-1 was trained on "enterprise-grade, clean and commercially licensed data," the company said, and the 35-billion-parameter model reportedly beat Anthropic's Sonnet 4.61 in a blind evaluation. Microsoft also said the model aligns with Anthropic Opus 4.6 on the SWE Bench Pro coding benchmark and is designed for multi-step tasks; it is in private preview on Microsoft Foundry.

The company also unveiled MAI-Code-1, described as "ultra-efficient" and "tuned for GitHub," which is coming to Copilot and VS Code on Tuesday. On the visual and audio fronts, MAI-Image-2.5 and a flash equivalent mark Microsoft's first models for text-to-image and image-to-image work and reportedly outperformed Nano Banana Pro on ELO.

microsoft, mai-thinking-1, reasoning model, mustafa suleyman, mai-code-1, copilot, vs code, microsoft foundry, mai-image-2.5, text-to-image