Microsoft AI MAI models replace OpenAI/Anthropic in M365 rollout

Microsoft says it is training sales staff to push in-house AI (MAI) instead of OpenAI and Anthropic. At Build 2026, the company unveiled seven Microsoft AI (MAI) models, including MAI-Thinking-1 (35B parameters, 256K context window). Microsoft’s pitch to enterprise buyers is clear: MAI models can match or beat GPT-5.5 and Anthropic Claude Opus 4.6 on key benchmarks and deliver up to 10x better cost efficiency in tuned workloads, especially for coding tasks such as SWE-Bench Pro. Microsoft also began a migration on July 7: it is moving selected Microsoft 365 apps (Excel and Outlook) away from OpenAI and Anthropic models onto its MAI stack. The company routes tens of thousands of prompts per week to internal models to reduce inference spending, targeting high-volume, lower-complexity work like spreadsheet formulas and email drafting, while keeping the Copilot user experience. Independent evaluations cited in the article support the push. In blind tests, human raters reportedly preferred MAI-Thinking-1 over Claude Sonnet 4.6. On coding benchmarks, MAI-Thinking-1 is described as competitive with Claude Opus 4.6. The 256K context window is positioned as a practical advantage for long documents (e.g., contracts and full earnings reports). For OpenAI and Anthropic, the risk is distribution and usage share. Azure and Microsoft 365 Copilot have historically driven demand for OpenAI and Claude models; diverting prompt volume to MAI could pressure their enterprise growth narratives.