MAI-Thinking-1
MAI-Thinking-1 is Microsoft AI's reasoning model for coding, math, and enterprise workflows. Microsoft describes it as a medium-sized sparse Mixture-of-Experts model with 35B active parameters, roughly 1T total parameters, a 256k token context window, function calling, and private preview availability in Microsoft Foundry.
MAI-Thinking-1 matters because Microsoft is presenting an in-house reasoning model stack rather than only distributing third-party models through Copilot and Azure. The page should help readers separate official model facts from early community reactions and benchmark comparisons.
Microsoft AI announced MAI-Thinking-1 on June 2, 2026 and describes it as a 35B-active, roughly 1T-total parameter sparse MoE model trained from the ground up on enterprise-grade, clean, commercially licensed data without third-party distillation. Microsoft claims strong software-engineering performance, high AIME scores, 256k context, function calling, Chat Completions API compatibility, private preview on Microsoft Foundry, and future MAI Playground availability. Hacker News and Reddit are useful only for demand and comparison language.
- Track Microsoft AI reasoning-model releases and availability.
- Compare Microsoft in-house models with Claude, OpenAI, DeepSeek, and Gemini reasoning models.
- Understand why model provenance, distillation claims, and enterprise data licensing matter.
- Decide whether to wait for Microsoft Foundry or MAI Playground access before evaluating the model directly.
Microsoft AI describes MAI-Thinking-1 as a reasoning model built for complex coding, math, and enterprise workloads. The official announcement says it is a 35B-active, roughly 1T-total parameter sparse MoE model, supports a 256k context window and function calling, and is compatible with Chat Completions-style workloads.
This should be treated as an entity explainer rather than a full GetLLMs model-directory record for now. Microsoft says MAI-Thinking-1 is available in private preview on Microsoft Foundry and will come to MAI Playground, but the stable public API IDs, pricing fields, and provider-level catalog details needed for a directory record are not yet complete in the evidence collected for this run.
The official page includes Microsoft-provided claims around SWE-Bench Pro, AIME 2025, AIME 2026, and blind human side-by-side preferences. Those claims are useful launch facts, but readers should treat them as official vendor evidence until independent model cards, provider listings, and third-party evaluations mature.
Microsoft AI coding model announced in the same MAI model family.
Coding-agent product often used as a comparison point for software-engineering model performance.
Recent model entity tracked for search and comparison demand around coding and reasoning models.
MAI-Thinking-1 FAQ
Page-level questions for MAI-Thinking-1.
What is MAI-Thinking-1?+
MAI-Thinking-1 is Microsoft AI's reasoning model for coding, math, and enterprise workloads. Microsoft describes it as a 35B-active sparse MoE model with roughly 1T total parameters, a 256k context window, function calling, and private preview availability on Microsoft Foundry.
Can I use MAI-Thinking-1 through a public API?+
Not as a fully stable public directory record in the evidence collected here. Microsoft says MAI-Thinking-1 is in private preview on Microsoft Foundry and will be available in MAI Playground soon, so API IDs, pricing, and provider availability should be checked again before production comparison.
Is MAI-Thinking-1 the same as MAI-Code-1-Flash?+
No. MAI-Thinking-1 is Microsoft AI's reasoning model, while MAI-Code-1-Flash is a smaller coding-focused model built for GitHub Copilot, VS Code, and the Microsoft stack.