100x Engineers
June 3, 2026
TL;DR
Microsoft launched seven proprietary AI models at Build 2026, including the reasoning model MAI Thinking One, reducing its reliance on OpenAI and enabling full control over its entire software stack.
“They're not just a distribution layer for OpenAI anymore. They're building the whole stack themselves.”
1. The Seven New Models Launch
Microsoft released seven AI models simultaneously at Build 2026, including MAI Thinking One for reasoning, MAI Code One Flash for coding, MAI Transcribe 1.5 for multilingual transcription, and MAI Voice 2 for text-to-speech with voice cloning.
2. Benchmark Performance and Capabilities
MAI Thinking One matched Claude Opus on software engineering benchmarks and beat Claude Sonnet in blind human testing across 1,276 real-world tasks, demonstrating competitive reasoning and coding capabilities.
3. Integration Across Microsoft's Ecosystem
Microsoft integrated these models directly into its owned platforms—GitHub, VS Code, Azure, Teams, and Office—creating a fully vertically integrated AI stack without external dependencies.
4. Strategic Shift Away from OpenAI Dependency
The $13 billion OpenAI partnership remains in place but is now optional, marking Microsoft's transition from distribution partner to independent AI stack provider with full technical control.