MICROSOFT JUST BUILT ITS OWN REASONING MODEL FROM SCRATCH — NO OPENAI, NO ANTHROPIC, NO SHORTCUTS Microsoft unveiled MAI-Thinking-1 at its Build 2026 developer conference, becoming the latest major technology company to build a frontier reasoning model entirely in-house, without relying on distillation from OpenAI, Anthropic, or any other third party. The model carries 35 billion active parameters, a 256,000-token context window, and was trained exclusively on commercially licensed data. Microsoft says MAI-Thinking-1 is optimized for complex multi-step instructions, long-context reasoning, and code generation. Internal benchmarks place its performance on par with Anthropic’s Claude Opus 4.6. The significance of the timing and framing is hard to miss. Microsoft has spent years as the most committed financial backer of OpenAI, pouring tens of billions into the partnership and integrating GPT models across virtually its entire product stack. MAI-Thinking-1 represents the company’s first credible signal that it intends to reduce that dependency. Build 2026 also introduced MAI-Code-1-Flash, a separate model tuned specifically for GitHub and VS Code that converts natural language descriptions into working code. Microsoft framed the broader set of seven new MAI models as proof that it can now compete at the frontier without borrowing someone else’s work. The message for OpenAI was clear enough: the partnership that made both companies rich may have a ceiling. Keywords: Microsoft MAI-Thinking-1, Microsoft AI model, MAI reasoning model, Microsoft OpenAI independence