
Microsoft launched seven AI models under the MAI brand, including MAI-Thinking-1 with 35 billion active parameters and 256k context window, trained from scratch on licensed data. The models cover reasoning, coding, image, voice and transcription. MAI-Thinking-1 is in private preview via Microsoft Foundry and blind human raters preferred it over Anthropic's Claude Sonnet 4.6. The AI stack runs across Windows, Azure and GitHub with a new server processor optimized for agent workflows. Microsoft emphasizes self-sufficiency over OpenAI dependence.
Tap to vote and see what everyone thinks.
New open-source voice model responds every 0.4 seconds
Summary by ByteBrief