Microsoft unveils MAI-Thinking-1 reasoning and MAI-Code-1 coding models

Microsoft officially announced a major expansion of its in-house MAI model family, developed by the Microsoft AI Superintelligence team.

At Build 2026, Microsoft today announced the significant expansion of its in-house AI model portfolio from the Microsoft AI Superintelligence Team. The new lineup includes the company’s first reasoning model, a coding model built for GitHub Copilot, and more.

Microsoft has been steadily growing its MAI model family over the past year. It previously announced MAI-Voice-1 and MAI-1-preview, followed by MAI-Transcribe-1 and MAI-Image-2 earlier this year. More recently, the company introduced MAI-Image-2.5, claiming improved text rendering, stylized illustrations, and commercial image quality compared to MAI-Image-2.

The new MAI-Thinking-1 is Microsoft AI’s first reasoning model. The company highlighted that this model was trained from scratch without distillation, using clean, commercially licensed, enterprise-grade data. It is a mid-sized model with 35 billion active parameters and a 128K context window. The model is designed to handle complex multi-step instructions, long-context reasoning, and code generation.

Without sharing any extensive benchmark comparisons, Microsoft noted in its blog post that independent raters preferred MAI-Thinking-1 over Anthropic’s Claude Sonnet 4.6 in blind testing. Microsoft also stated that the model matches Claude Opus 4.6 on coding tasks in SWE-bench Pro. MAI-Thinking-1 is now available in private preview through Microsoft Foundry.

Microsoft's recently announced MAI-Image-2.5 and its new flash variant are now available for developers via Microsoft Foundry. As we reported last month, the MAI-Image-2.5 model beats Google’s Nano Banana Pro on Arena’s latest text-to-image leaderboard. This model is already available in PowerPoint and is rolling out to OneDrive.

In April, Microsoft announced MAI-Transcribe-1 with support for speech-to-text transcription across the 25 most-used languages based on its product usage data. Today, the company introduced MAI-Transcribe-1.5, an updated version of its speech-to-text model that delivers state-of-the-art accuracy across 43 languages. It was also announced that streaming support is coming soon to this model.

Additionally, following the general availability of the MAI-Voice-1 speech generation model in April, Microsoft unveiled MAI-Voice-2 and a flash variant today, featuring support for more than 15 additional languages and new voice options.

Finally, Microsoft announced MAI-Code-1, an inference-efficient coding model tuned for GitHub workloads. The model is now available in GitHub Copilot and Visual Studio Code. While benchmark results for MAI-Code-1 were not provided, this announcement is significant because Microsoft has traditionally relied on models from OpenAI and Anthropic to power coding experiences in GitHub Copilot.

Beyond Microsoft Foundry and its own products, the company announced that its MAI models will also become available through Fireworks AI, Baseten, and OpenRouter. In addition, Fireworks AI is now generally available within Microsoft Foundry.