Microsoft's Superintelligence team has launched MAI-Code-1-Flash, the company's first in-house coding model purpose-built for the GitHub Copilot harness. Announced on June 2, 2026, the 5-billion-parameter model is rolling out to every Copilot tier — Free, Pro, Pro+ and Max — through the VS Code model picker and the new Auto router, with no additional setup required.
Built end-to-end for production Copilot workflows
Microsoft says MAI-Code-1-Flash was trained from March to May 2026 on "clean and appropriately licensed data" and has a 256K-token context window. Crucially, it was trained directly against the GitHub Copilot harness developers use in production rather than against synthetic benchmark suites, with checkpoints evaluated on real-world software engineering tasks like refactoring, repository QA and telemetry-grounded code edits.
Beats Claude Haiku 4.5 on every benchmark Microsoft tested
The headline number: a +16-point lead over Anthropic's Claude Haiku 4.5 on SWE-Bench Pro (51.2% vs 35.2%), with higher pass rates on SWE-Bench Verified, SWE-Bench Multilingual and Terminal Bench 2 as well. Microsoft also reports up to 60% fewer tokens used to solve harder problems on SWE-Bench Verified, thanks to an "adaptive solution length" technique that lets the model stay concise on trivial requests and spend more reasoning budget on complex ones.
Part of a wider MAI push to reduce OpenAI dependency
MAI-Code-1-Flash is one of seven in-house MAI models the team has shipped or previewed since Build 2026, alongside MAI-Thinking-1 (reasoning), MAI-Image-2.5 (image generation) and MAI-Transcribe-1.5 (speech). Microsoft's broader strategy is to give Copilot a model bench it owns end-to-end, reducing latency and unit cost for Copilot's hundreds of millions of users and lowering its reliance on external frontier labs.
What it means for developers
For Copilot users, the rollout means cheaper, faster autocomplete and agentic actions on routine tasks, with the Auto picker dynamically routing harder problems to larger models. For Microsoft, it's another step toward the model-ownership posture that has reshaped its chip and infrastructure partnerships with NVIDIA and TSMC, and it lands the same week that NVIDIA shipped its own Cosmos 3 open foundation model for physical AI.
Reporting based on coverage from Microsoft AI, CNBC, ChatForest and Simon Willison's blog.