Microsoft's MAI-Thinking-1 matches Sonnet 4.6 in blind evaluations

Microsoft kicked off Build 2026 with the announcement of MAI-Thinking-1, a new flagship reasoning model built by Mustafa Suleyman's Microsoft AI division. The model is specifically engineered for complex tasks that require multi-step reasoning, software engineering, and research synthesis. Early benchmarks indicate that MAI-Thinking-1 matches leading models on key software engineering evaluations, and importantly, it also achieves human preference parity with Anthropic's Sonnet 4.6 in blind evaluations—a strong signal that Microsoft's reasoning capabilities are now competitive with the best in the industry.

The launch positions Microsoft as a serious contender in the reasoning model space, challenging OpenAI's o-series models and Anthropic's Claude. MAI-Thinking-1 is expected to be integrated across Microsoft's product ecosystem, including GitHub Copilot and Azure AI Studio, giving developers and enterprises a powerful new tool for coding, analysis, and decision-making. By matching Sonnet 4.6 on human preference and leading benchmarks, Microsoft is signaling that it can deliver frontier-level reasoning without relying on external partners.