Microsoft has stepped up its in-house artificial intelligence game with the unveiling of MAI-Thinking-1, a medium-sized model that competes with industry leaders on key software engineering benchmarks. The new model is part of a broader set of seven announcements at Build 2026, including advances in image generation, transcription, voice AI and coding tools.
The company claims to have trained MAI-Thinking-1 from scratch using clean data, setting it apart from its previous reliance on third-party models. Meanwhile, other highlights include the faster MAI-Transcribe-1.5, which is five times quicker than rivals, and the multilingual capabilities of MAI-Voice-2.
The new coding model, MAI-Code-1-Flash, integrates seamlessly with popular development tools like GitHub Copilot and Visual Studio Code, promising to boost productivity for developers.
In an era where AI is increasingly integral to tech innovation, Microsoft’s investment in homegrown models could signal a shift towards more proprietary solutions. However, the real test will be how these models perform in the field compared to their open-source counterparts.
