Microsoft Launches MAI-Thinking-1: A 1T Parameter Reasoning Model Optimized for Cost-Efficiency

finbarrtimbers · x · 2026-08-13

Microsoft has launched its in-house reasoning model, MAI-Thinking-1, in preview on Azure AI Foundry. The model is a sparse Mixture-of-Experts (MoE) Transformer with a total of 1T parameters and 35B active parameters.

Designed as a reasoning model, it generates an internal chain of thought before emitting the final response, adaptively allocating reasoning effort based on prompt complexity. Microsoft highlights that it delivers top-tier reasoning at the best price-to-performance ratio in its weight class. The team is currently seeking public feedback, particularly on its tool-use capabilities.

Related event: Microsoft Launches First In-House Reasoning Model MAI-Thinking-1(3 posts)→

Original post →

More from Models

Models channel →