Microsoft has expanded its in-house AI model lineup with four new models, including its first reasoning model. Microsoft is developing its own AI models to lessen dependence on external providers.
New models expand Microsoft's AI portfolio
The models were developed by Microsoft's AI Superintelligence Team. MAI-Thinking-1 is a reasoning model trained from scratch with 35 billion active parameters and a 128K context window. MAI-Code-1 is a coding model tuned for GitHub Copilot and Visual Studio Code.
MAI-Thinking-1 handles complex multi-step instructions and long-context reasoning. Independent raters preferred it over Anthropic's Claude Sonnet 4.6 in blind testing, according to Microsoft. It also matches Claude Opus 4.6 on coding tasks in SWE-bench Pro.
MAI-Thinking-1 leads with reasoning capabilities
MAI-Transcribe-1.5 updates the speech-to-text model with support for 43 languages, with streaming support coming soon. MAI-Voice-2 adds over 15 languages and new voice options.
MAI-Thinking-1 is available in private preview through Microsoft Foundry. The MAI models will also be available through Fireworks AI, Baseten, and OpenRouter.
Microsoft had previously released earlier versions of MAI models including MAI-Voice-1, MAI-1-preview, MAI-Transcribe-1, and MAI-Image-2. The new models significantly expand the MAI family.
The confirmed details include the four new models, their key specs, availability, and performance claims. Pricing has not been announced.
Discussion
0 comments
Log in to join the thread with a thoughtful take, question, or correction.