Microsoft's MAI-Thinking-1 reasoning model claims parity with Claude Opus 4.6 on coding

Microsoft AI released MAI-Thinking-1 on August 12, 2026, its first reasoning model trained from scratch, and said the system matches Anthropic’s Claude Opus 4.6 on a key coding benchmark despite running far fewer active parameters.

The launch is part of Microsoft’s push to build frontier reasoning in-house rather than lean on partners, and lands as the company diversifies beyond its OpenAI relationship. Microsoft says it introduced the model with no distillation from third-party frontier models.

MAI-Thinking-1 is a sparse mixture-of-experts model with 35 billion active parameters — roughly 1 trillion in total — and a 256,000-token context window, trained on public and licensed data, the company said.

On benchmarks, Microsoft says the model scores 97.0% on AIME 2025 and 94.5% on AIME 2026, two competition-math tests, and matches Claude Opus 4.6 on SWE-Bench Pro, a software-engineering benchmark. It also says human raters preferred the model’s outputs over those of Claude Sonnet 4.6 in blind side-by-side testing across 1,276 tasks.

Those figures are Microsoft’s own and have not been independently verified. Vendor benchmark results routinely flatter the vendor’s model, and ‘human raters preferred’ comparisons depend heavily on how the test was run and who ran it.

MAI-Thinking-1 is available in public preview through Microsoft Foundry, the company’s model catalog. Broader availability and pricing will show whether the in-house model can hold up against Claude and other frontier systems outside a controlled benchmark.