MAI-1-preview
Microsoft's first end-to-end trained foundation model — MoE, publicly tested on LMArena.
About
MAI-1-preview is Microsoft AI's first foundation model trained fully in-house, announced August 28, 2025 by Microsoft AI CEO Mustafa Suleyman. It is a mixture-of-experts language model trained on roughly 15,000 NVIDIA H100 GPUs, designed for instruction following and everyday queries — and, pointedly, to reduce Microsoft's dependence on OpenAI models for select Copilot text workloads.
Microsoft put it into public testing on LMArena to gather feedback and began rolling it into select Copilot text features in the weeks after launch, with API access available by application. Publicly disclosed specs are thin: no context window, pricing, or benchmark figures have been published, so treat every capability claim as provisional until Microsoft or independent testing says otherwise.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
not published; model is under public testing on LMArena.
Independent: no published third-party results found.
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 6/10
Announcement confirmed by Microsoft AI's own channels and independent press (DeepLearning.AI, Insider Monkey) with consistent facts. Kept at metrics tier because: (1) almost no public specs — no benchmarks, context window, or pricing disclosed; (2) availability is limited to preview channels with API by application only.