MAI-Thinking-1 parameter configuration, around 128K context + MoE architecture, Excel version reduces costs by 10 times and aligns with GPT-5.4, the enterprise market is about to change.

Microsoft unveiled the MAI native model family at Build 2026: 7 models, zero distillation, data compliance, and a goal of human-centric superintelligence. The flagship MAI-Thinking-1 features 35 billion parameters, a 128K context window, and MoE. There are also MAI-Code-1-Flash, MAI-Image-2.5, MAI-Transcribe-1.5, and MAI-Voice-2, which can integrate with Copilot/VS Code and support weight fine-tuning. The Maia 200 chips improve computing efficiency by 1.4 times. With Frontier Tuning, enterprises can customize models in “training gyms.” The Excel version aligns with GPT-5.4 and cuts costs by 10 times. Mayo Clinic will jointly develop clinical reasoning large models—first for in-hospital use, and then open them to Azure Foundry.