Microsoft's MAI-Thinking-1 reasoning model beats Anthropic Sonnet 4.61 in blind tests

Microsoft's MAI-Thinking-1, its first reasoning model, outperformed Anthropic's Sonnet 4.61 in independent blind tests and aligns with Anthropic Opus 4.6. Trained on enterprise-grade, clean, and commercially licensed data, MAI-Thinking-1 addresses copyright concerns amid active lawsuits. The 35-billion-parameter model was evaluated by independent reviewers and demonstrated superior performance in reasoning tasks. This marks Microsoft's first dedicated reasoning AI, part of seven new AI models unveiled at Build. The model's humanist superintelligence framing emphasizes ethical AI development and customer trust, positioning it as a competitive alternative to existing reasoning models.