Microsoft Breaks From OpenAI, Ships Seven MAI Models at Build 2026
Microsoft ships seven in-house AI models at Build 2026, from a frontier reasoning model to voice and transcription, all trained without OpenAI data
- Microsoft launched 7 in-house AI models at Build 2026, covering reasoning, coding, image, transcription, and voice — all trained without OpenAI data or distillation.
- MAI-Thinking-1 is a 35B-active / ~1T-total parameter MoE reasoning model scoring 97% on AIME 2025 and 53% on SWE-Bench Pro, matching Claude Opus 4.6.
- MAI-Code-1-Flash (5B params) beats Claude Haiku 4.5 by 16 pts on SWE-Bench Pro using 60% fewer tokens, and is now live in GitHub Copilot and VS Code.
- MAI-Image-2.5 Flash is priced at $1.75/1M input tokens; MAI-Transcribe-1.5 at $0.36/hour; MAI-Voice-2 at $22/1M characters — all available via Microsoft Foundry.
- MAI-Thinking-1 is in private preview on Foundry; models are also available on Fireworks AI, Baseten, and OpenRouter outside of Azure.
- Microsoft introduced Frontier Tuning, letting enterprises train MAI models on their own workflow data — a tuned Excel model reportedly matches GPT 5.4 at 10x lower cost.
For three years, Microsoft's AI products ran almost entirely on OpenAI models. At Build 2026, Mustafa Suleyman walked on stage and changed that. Microsoft launched seven new in-house AI models at Build 2026, marking the company's most significant push yet to build its own frontier AI stack independent of OpenAI. The lineup spans reasoning, coding, image generation, transcription, and voice , and every single model was built from scratch, with no shortcuts.
Zero distillation, clean slate
The phrase that keeps appearing in Microsoft's documentation is "zero distillation." Most smaller models are trained by learning to imitate a larger, more capable model , a technique called knowledge distillation. Most smaller models are trained by learning to imitate a larger, more capable model. MAI-Thinking-1 was trained without doing this. Microsoft argues that distilled models are fundamentally bound to the design choices of their teacher model and struggle to generalize to new situations. Training from scratch on their own data means the model has to genuinely learn reasoning rather than mimicking it.
Every new MAI model is trained from scratch using "clean and appropriately licensed data, without distillation from third-party models" , language that directly addresses the intellectual property concerns that have accompanied third-party model licensing. Microsoft is also co-designing models with its own Maia 200 silicon, and says it is already seeing a 1.4x efficiency boost from that work.
The flagship: MAI-Thinking-1
MAI-Thinking-1 is based on a sparse mixture-of-experts architecture with 35 billion active parameters out of approximately one trillion total parameters and a 256,000-token context window. The MoE architecture (Mixture of Experts) works by selectively activating only the subset of the model's parameters needed for each request , so you get the capacity of a trillion-parameter model at the inference cost of a 35B one.