Microsoft Ships MAI-Thinking-1 and MAI-Code-1-Flash, Its First Foundation Models Without OpenAI
At Build 2026 Microsoft unveiled an in-house reasoning model and a Copilot-tuned coding model, both trained from scratch on commercially licensed data with no distillation from third-party models — the clearest signal yet that Redmond is decoupling its model stack from OpenAI.
Microsoft used the June 2 Build 2026 keynote to introduce MAI-Thinking-1, a 35B-active-parameter reasoning model with a 128K context window, and MAI-Code-1-Flash, an inference-efficient coding model that began rolling out the same day to all GitHub Copilot tiers including the free plan. The company stated explicitly that both models were trained from scratch on enterprise-grade, commercially licensed data without distillation from any third-party model — language that is hard to read as anything other than a fence around the OpenAI relationship.
The benchmark numbers are the load-bearing part of the story. MAI-Thinking-1 posts 97.0% on AIME 2025 and 94.5% on AIME 2026, and Microsoft says it matches Claude Opus 4.6 on SWE-Bench Pro coding tasks. MAI-Code-1-Flash is the more pointed claim: a 51.2% to 35.2% lead over Claude Haiku 4.5 on SWE-Bench Pro and up to 60% fewer tokens to solve the same SWE-Bench Verified problems. Those are not parity claims; they are price-performance claims aimed directly at the cheap-tier coding workloads that drive Copilot's seat-license economics.
The strategic significance is that Microsoft now has a credible internal substitute for the OpenAI capacity it has been buying for years, deployed inside its largest developer product, on the same day the model was announced. That does not mean Microsoft will stop paying OpenAI — the $50B+ commitments are still in force — but it does mean Redmond has retired the argument that it could not, technically, do so. The MAI launch should be read in the same week as Anthropic's S-1: both events compress the time horizon in which OpenAI's distribution dependence becomes a priced risk.
Microsoft AI — Introducing MAI-Code-1-Flash → · CNBC — Microsoft unveils new AI models to lessen reliance on OpenAI and lower costs for developers → · Neowin — Microsoft unveils MAI-Thinking-1 reasoning and MAI-Code-1 coding models →