mai

The 2026 Microsoft Build Conference is officially kicking off, and Microsoft is rolling out its in-house MAI model family with 7 native models across inference, programming, image processing, speech, and transcription. The flagship model, MAI-Thinking-1, boasts 35 billion active parameters and a 128K context window, hitting levels comparable to Claude Opus 4.6 on the SWE-bench Pro. They also launched cutting-edge Frontier Tuning services, a Web IQ search engine, and the next-gen topology quantum chip, Majorana 2, which boosts qubit reliability by 1000 times, pushing the timeline for scalable quantum computers to 2029.

Why it matters: This marks Microsoft's first fully in-house AI model stack (no third-party distillation involved), while advancing their quantum computing + AI Agent platform, signaling a new phase in the tech giant's AI competition, where hardware and software are integrated as one.

#微软 #MAI #AI #量子计算 #Build2026