Microsoft launches MAI-Thinking-1 — its own reasoning model — in public preview on Azure AI Foundry

At Build 2026 on June 2, 2026, Microsoft launched MAI-Thinking-1 — a mid-weight large language model for chat and reasoning — in public preview through Azure AI Foundry. It was one of four new MAI models entering the Foundry catalog, part of what Microsoft's official Build news framed as seven new MAI models overall. The launch is the latest sign of Microsoft building out first-party models alongside the OpenAI models it has historically deployed across its products.

What MAI-Thinking-1 is. Microsoft's official description is concise: MAI-Thinking-1 is "a mid-weight large language model for chat and reasoning," available to try in the Foundry catalog. Microsoft did not disclose model size, training compute, or benchmark scores in its Build announcement. Some Build coverage reported that MAI-Thinking-1 was trained from scratch without distillation from third-party frontier models such as OpenAI's — a notable claim if accurate, but one that does not appear in Microsoft's own published Build materials as of writing. We report the official framing — a first-party reasoning model in public preview — and flag the "no distillation" detail as reporting Microsoft has not confirmed.

The other MAI models. Three additional MAI models entered public preview alongside MAI-Thinking-1: MAI-Image-2.5 (an updated image generator with image-to-image editing), MAI-Transcribe-2 (a speech-to-text model with speaker diarization and content biasing), and MAI-Voice-2 (a multilingual text-to-speech model with voice cloning). Together they cover the four core generative modalities — text, image, transcription, and voice. Pricing for the new models was not announced at Build; the Foundry catalog listing is the access path for developers.

The broader Foundry positioning. The MAI family is part of Microsoft's wider "AI optionality" pitch at Build: within a single Azure AI Foundry deployment, enterprise customers can route workloads through Microsoft's own MAI models, OpenAI, or third-party model families that Foundry supports — including Claude, Llama, DeepSeek, and Mistral — under one governance and billing stack. Foundry also shipped platform updates at Build: Foundry IQ Knowledge Bases reached general availability, Voice Live became generally available for prompt agents, and hosted agents in Foundry Agent Service are expected to reach general availability by early July 2026.

For developers building on Azure. MAI-Thinking-1 is a new option in the Foundry catalog for reasoning and chat tasks, governed and billed within the same Azure stack teams already use. Public preview status means it is available to evaluate but not production-hardened — treat it as an evaluation candidate, not an immediate production replacement for an established model choice. The Foundry hosted-agents runtime (expected GA in early July 2026) is the intended deployment layer for production agentic use.