Microsoft Unveils MAI-Thinking-1 and New AI Model Suite at Build 2026
At Build 2026, Microsoft doubled down on in-house AI with MAI-Thinking-1, a “medium-sized” reasoning model developed from the ground up on clean data (no distillation), matching leading models on key software engineering benchmarks. It also rolled out MAI-Transcribe-1.5 (5× faster), MAI-Voice-2 (+15 languages, flash “coming soon”), MAI-Image 2.5, and MAI-Code-1-Flash (inference-efficient, integrated into GitHub Copilot and VS Code).
Microsoft announced a new lineup of in-house AI models at its Build 2026 conference, headlined by MAI-Thinking-1, the company’s first advanced reasoning model. The announcement, made in June 2026, marks a continued shift toward independent AI development for Microsoft.
MAI-Thinking-1 is described by Microsoft as a “medium-sized model” that “matches leading models” on “key” software engineering benchmarks. The company says it “trained it from the ground up on clean data, without distillation from third-party models.” Microsoft began developing in-house models only last year, having previously relied on OpenAI’s models. The two companies have since renegotiated their partnership to loosen ties.
Alongside MAI-Thinking-1, Microsoft introduced several other models spanning different capabilities. MAI-Image 2.5 and its flash variant support text-to-image generation and image editing. MAI-Transcribe-1.5 is described as “five times faster than competing models.” MAI-Voice-2 adds 15 new languages and expanded voice options, with a flash version described as “coming soon.” A new coding model, MAI-Code-1-Flash, is positioned as “inference-efficient” and has been integrated into GitHub Copilot and Visual Studio Code.
The breadth of the release — covering reasoning, image generation, transcription, voice, and coding — suggests Microsoft is moving to build out a broad portfolio of proprietary AI capabilities. For developers and enterprise users already working within Microsoft’s ecosystem, the integration of MAI-Code-1-Flash into GitHub Copilot and Visual Studio Code may offer a more immediate point of contact with the new models.