7 Microsoft AI models aim at OpenAI and Anthropic

Microsoft unveiled seven in-house AI models at Build 2026 to reduce reliance on OpenAI and Anthropic.

Microsoft unveiled seven in-house AI models at its Build 2026 conference in San Francisco on 3 June, marking a sharper push to control more of its AI stack.

What changed

The headline model is MAI-Thinking-1, Microsoft’s first reasoning model. It was trained from scratch on commercially licensed data, not distilled from another company’s system.

Microsoft said MAI-Thinking-1 is built for multi-step instructions, long-context reasoning and code generation. The model has 35 billion active parameters and a 256,000-token context window.

Microsoft also launched MAI-Code-1-Flash, a coding model that turns text prompts into source code for apps and websites. It is rolling out in GitHub Copilot and Visual Studio Code.

• Microsoft says MAI-Thinking-1 beat OpenAI’s GPT-5.5 on quality in a McKinsey tuning test.
• The company says that result came with up to 10x better cost efficiency, using public pricing data.
• In blind tests run by Surge, MAI-Thinking-1 was preferred over Anthropic’s Claude Sonnet 4.6.
• Microsoft says it matches Claude Opus 4.6 on coding benchmarks.

Satya Nadella said companies should move from “consuming a frontier model” to participating in it. Mustafa Suleyman, who leads Microsoft AI, framed the launch as a step toward more model independence after years of heavy spending on outside AI firms.

Why it matters

The new models let Microsoft run more AI workloads on Azure instead of paying third-party model providers. That could lower inference costs and give the company more room to price Copilot and other developer tools aggressively.

For developers, the immediate impact is broader access inside Microsoft’s own products. For the market, it is another sign that Microsoft wants to compete with OpenAI and Anthropic even as it remains one of the biggest backers of both.

The timing also matters: Microsoft’s biggest AI bets are heading toward IPO plans, which could change how much leverage the company has over its partners. The question now is whether Microsoft can turn in-house models into a real product edge, not just a cost-saving move.