Microsoft AI's mixture-of-experts reasoning model, in private preview via Microsoft Foundry.
Updated EU status checked Last shown
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
MAI Thinking 1 is a reasoning model from Microsoft AI (MAI), listed on microsoft.ai among its foundational models. Microsoft describes it as built for complex problems, with competitive reasoning and top SWE-Bench Pro results at a mid-weight price.
According to a YouTube creator summary, it is a mixture-of-experts model with about 1 trillion total parameters and 35B active. It reportedly outperforms Claude Sonnet 4.6 on several benchmarks while remaining behind the top frontier models. These figures come from secondary coverage, not from the scraped Microsoft pages.
It is reportedly offered as a private preview in Microsoft Foundry, so access is limited to approved customers. Weights are not published, and no public documentation or paper was found.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Microsoft's 1T-parameter mixture-of-experts thinking model with 35B active, outperforming Sonnet 4.6 on several benchmarks but behind top models. In private preview in Microsoft Foundry.
Closest neighbours by meaning, not just by category.
Model · Microsoft
Microsoft AI's in-house MAI model family covering reasoning, code, speech, transcription, image and cybersecurity.
Model · Microsoft
Microsoft AI's pre-trained foundation model with about 80B active parameters; post-training unfinished and not yet available.
Model · Microsoft
Microsoft AI's text-to-speech voice model with expressive output across multiple languages.
Model · Microsoft
Microsoft's coding-focused AI model in the MAI family, announced with few public details.
Model · Microsoft
Microsoft's coding-focused language model, positioned as a faster, more token-efficient alternative to Claude Haiku 4.5.
Model · OpenAI
OpenAI's first reasoning model series, trained with reinforcement learning to think before answering; now deprecated in the API.
