Top AI Product

title: “Microsoft MAI-Thinking-1 outscores Claude Sonnet 4.6 in blind evals — trained without OpenAI’s data”
date: 2026-06-03
tags: [model, microsoft, reasoning, api]

Microsoft shipped its first fully self-built reasoning model at Build 2026, and the signal is hard to miss.

MAI-Thinking-1 hits 97% on AIME 2025 and 94.5% on AIME 2026. In blind human evaluations on software engineering tasks, it beat Claude Sonnet 4.6. The model is mid-sized, not frontier-scale, but the benchmark scores are competitive with the top tier.

What makes it different

Microsoft trained MAI-Thinking-1 entirely on enterprise-grade clean data — no distillation from third-party models, including OpenAI. That’s the part worth paying attention to. Every major lab has been training on synthetic outputs from competitors. Microsoft is breaking from that playbook and owning the full training pipeline independently.

The strategic read: Microsoft has been OpenAI’s largest investor and customer. Building a parallel model that doesn’t depend on OpenAI data is a deliberate hedge. It’s not replacing the partnership — it’s reducing dependency.

API access via Microsoft Foundry

MAI-Thinking-1 is available in private preview through the Microsoft Foundry API. The entry point is aimed at enterprise developers building coding tools, complex reasoning pipelines, or software automation workflows. If the benchmark numbers hold in production, it’s worth getting on the waitlist.

You Might Also Like

• 13b Into Openai yet Microsoft Copilot Cowork Runs on Anthropic Claude
• Microsoft mai Models mai Transcribe 1 mai Voice 1 mai Image 2 are Live Redmonds ai Independence Starts now
• Anthropics Claude Code Openclaw Restriction Policy 200 Month Users Push Back Openai Opens its Doors
• 26 Engineers 20m Arcee ai Trinity Large Thinking Scores Within 2 Points of Claude Opus
• Snap Specs Signs Multiyear Qualcomm Deal Consumer ar Glasses With Openai and Gemini Ship h2 2026