Microsoft presented seven MAI models at its Build 2026 conference on June 2, 2026, in San Francisco, before unveiling six additional models at Ignite 2026. The lineup covers coding, cybersecurity, image generation, transcription and voice. Its flagship model, MAI-Thinking-1, has 35 billion active parameters, a 256K context window and a score of 97% on the AIME 2025 mathematics benchmark. MAI-Code-1-Flash targets multi-step programming tasks, while MAI-Transcribe-1.5 supports 43 languages and MAI-Voice-2 can clone voices in 15 languages. The models are integrated with GitHub Copilot, Visual Studio Code and Microsoft Foundry, and Microsoft says they were trained on licensed commercial data without using distillation from third-party models.
Source: Read the original article


