Noesion Radarnoesion.ai
Release2026-08-12

Microsoft released MAI-Thinking-1, its own reasoning model, in public preview on Microsoft Foundry. The company says it was trained from scratch on traceable data without copying outputs from other labs, matches Claude Opus 4.6 on a software engineering benchmark, and was preferred to Claude Sonnet 4.6 by paid human raters. All benchmark and preference figures are Microsoft's own, and the model is still in preview.

What changed

Microsoft's in-house MAI line had no dedicated step-by-step reasoning model of this size.

What it unlocks

Running a reasoning model with a 256k-token window and function calling through Microsoft Foundry, with Chat Completions API compatibility.

  • 35B active, ~1T total parameters
  • 256k token context window
  • 97.0% AIME 2025, 94.5% AIME 2026
  • 1,276 tasks in human comparison

What you need to act on it

  • Microsoft Foundry access
  • public preview

Send this to someone who needs it

Shares the story and its sources. Nothing about you.

What does this mean for your job?

This is the story as everyone gets it. Once a week we send you the version written for your role — what changed, why it matters for the work you actually do, and one thing to try. Free while we tune it.