Microsoft launches seven in-house 'MAI' models at Build 2026, led by 35B-active MAI-Thinking-1 trained without OpenAI data

At its Build 2026 developer conference in San Francisco on June 2, 2026, Microsoft formally launched its homegrown MAI model family — seven models in total — its most explicit move yet to reduce dependence on OpenAI. The flagship is MAI-Thinking-1, Microsoft AI's first reasoning model: a sparse mixture-of-experts design with 35 billion active parameters (roughly one trillion total) and a 256,000-token context window.

Microsoft stresses it was trained 'entirely on clean, commercially licensed data without distillation from any third-party model.' On benchmarks it posted 97.0% on AIME 2025 and 94.5% on AIME 2026; on SWE-Bench Pro (~52.8%) Microsoft says it matches Anthropic's Claude Opus 4.6 on coding and that blind human raters preferred it over Claude Sonnet 4.6. The other six models: MAI-Code-1-Flash (a 5B coder hitting ~51% SWE-Bench Pro, beating Claude Haiku 4.5 at lower cost), MAI-Image-2.5 and a Flash variant, MAI-Transcribe-1.5 (43-language transcription), and MAI-Voice-2 plus a Flash variant (15+ languages).

Microsoft AI CEO Mustafa Suleiman framed the strategy around 'long-term self-sufficiency,' claiming tuned MAI models beat OpenAI's GPT-5.5 on quality for specific enterprise tasks while delivering a 'tenfold reduction in costs.' Because Microsoft owns both the models and the Azure compute, it can avoid third-party royalties and pass savings to developers. The launch marks a strategic inflection: after years as OpenAI's largest backer and distribution partner, Microsoft is now competing head-on at the foundation-model layer.