Microsoft Built Its Own AI Models — And They're Good

Microsoft just did something that changes the AI landscape for developers: at Build 2026, they launched seven in-house AI models trained completely from scratch — no distillation from OpenAI, no data borrowed from Anthropic. This is a big deal because Microsoft has poured billions into both companies.

What they shipped

The flagship is MAI-Thinking-1, a 35B parameter reasoning model with a 256K context window. In blind evaluations, it outperformed Claude Sonnet 4.6 and matched Claude Opus 4.6 on coding benchmarks. Microsoft also launched MAI-Code-1-Flash, a coding model already rolling out in GitHub Copilot and VS Code.

The strategic move here isn’t just capability — it’s cost. Running their own models on Azure means Microsoft can bypass third-party API fees entirely and pass those savings to developers building on their platform.

What this means for builders

If you’re building on Azure or using GitHub Copilot, these models are coming to your workflow whether you opt in or not. More competition at this level is good — it pushes pricing down across the board.

Test MAI-Thinking-1 in the MAI Playground once it opens and benchmark it against your current stack before committing either way.