Microsoft releases seven in-house frontier models including MAI-Thinking-1 (35B), claims parity with Opus 4.6 on coding, trains on clean licensed data

At Build 2026, Microsoft announced a family of seven new models including MAI-Thinking-1, a 35B parameter reasoning model trained from scratch on enterprise-grade, commercially licensed data with zero distillation. Independent raters prefer it to Sonnet 4.6 on blind tests; it matches Opus 4.6 on SWE Bench Pro for coding. Image models (MAI-Image-2.5) rank #3 and #2 on Arena leaderboards; transcription, voice, and code models also released. All are positioned to reduce enterprise training-data IP liability and commoditize access to frontier capability outside the few dominant vendors. For financial institutions tracking model concentration and data-cost risk, this signals Microsoft's intent to reduce dependency on OpenAI and expand the model supply surface for governed enterprise workloads.