MAI-Thinking-1
Language modelCard summarizedBuilt for the complex problems that matter most
Public previewSource: MAI-Thinking-1: Building a Hill-Climbing Machine (technical report) (109 pp)
Use it via API
Where to call it
- Microsoft Foundry / Azure
MAI-Thinking-1 - MAI Playground
mai-thinking-1-latest
Good for
- Enterprise deployments that need 256K context, function calling and complex instruction following
- Coding workflows: reading code, editing files, running tests, fixing bugs and recovering from mistakes
- Quantitative reasoning such as financial modeling, statistical analysis, market sizing and forecasting
Overview
MAI-Thinking-1
Built for the complex problems that matter most
- Cost
- $2.00 in
- Context
- 256K
- Input / output
- $2.00 / $8.00
- Max output
- 64K
- Knows up to
- July 2025
- Released
- 12 Aug 2026
Microsoft AI's first reasoning model: a sparse mixture-of-experts model with about 35B active and 1T total parameters, trained from scratch on clean, licensed data without distillation from third-party models. Microsoft places it in the competitive range of popular LLMs rather than leading. It reports 52.8% on SWE-Bench Pro and 97.0% on AIME 2025, and roughly Claude Sonnet 4.6 level on general benchmarks. It runs on Microsoft Foundry as a text-only preview with a 256K context. p.1–2, §4.1
MAI-Thinking-1- Input / output per 1M
- $2.00 / $8.00
- Context window
- 256K
- Max output
- 64K
- Knows up to
- July 2025
- Reasoning
- Always on; effort adapts to the prompt
- Weights
- Closed
About this card. A model card is the maker's own account of a model: what it's for, how it was trained and tested, and where it falls short. This page condenses Microsoft AI's MAI-Thinking-1: Building a Hill-Climbing Machine (technical report). Every claim carries its source page.