At its Build conference on June 2, Microsoft introduced a family of in-house AI models under the MAI name — seven systems spanning reasoning, coding, image generation, voice, and transcription.
For developers, the headline is MAI-Code-1, which Microsoft describes as an "inference efficient coding model tuned for GitHub." It is available now in Copilot and in VS Code, putting a Microsoft-built model directly inside the tools millions of developers already use.
The flagship is MAI-Thinking-1, billed as Microsoft's first reasoning model. The company says it was trained from scratch on commercially licensed data, runs 35 billion active parameters with a 256,000-token context window, and is in private preview on its Foundry platform.
Microsoft's quality claims are striking, and they are Microsoft's own. The company says independent raters prefer MAI-Thinking-1 to Anthropic's Sonnet 4.6 and that it matches Opus 4.6 on the SWE-Bench Pro coding benchmark. Those figures have not been independently verified, and vendor benchmark claims rarely survive contact with outside testing unchanged.
The rest of the lineup fills out the stack: MAI-Image-2.5 for image generation, MAI-Voice-2 for speech, and MAI Transcribe 1.5 for transcription across dozens of languages. Microsoft says the models will also be available through Fireworks AI, Baseten, and OpenRouter.
The framing matters as much as the models. CNBC reported that the launch is part of Microsoft's effort to reduce its reliance on OpenAI and to lower costs for developers building on its platforms.
That is the real story. Microsoft is OpenAI's largest backer and has built much of Copilot on OpenAI's models. Shipping its own competitive models — and slotting one into Copilot and VS Code — is a hedge against that dependence and a shift in who holds leverage in the partnership.
Why it matters: the company that helped make OpenAI a household name is now selling rivals to it, inside its own developer tools. For developers, it means more choice and likely lower prices. For the broader map of power in AI, it is one more sign that the big platforms want to own the models, not just rent them.