MAI's 7 New Models

Microsoft AI announced a family of seven new models developed in-house, spanning image, voice, transcription, coding, and reasoning. The company describes the MAI model family as a multimodal ecosystem designed to work across real-world tasks. The models include MAI-Thinking-1, a medium-sized flagship reasoning model that matches leading models on key software engineering benchmarks and demonstrates advanced mathematical reasoning. MAI-Code-1-Flash is an inference-efficient agentic coding model integrated into GitHub Copilot, VS Code, and the Microsoft stack, with 5 billion active parameters. MAI-Image-2.5 and its Flash variant support text-to-image and image editing. MAI Transcribe-1.5 is a transcription model with SOTA accuracy, five times faster than competing models, and support for domain-specific terminology across 43 languages. MAI-Voice-2 generates natural-sounding speech across 15 languages and can adapt to a voice from a short sample. The models are distributed on Foundry and optimized for first-party products, and will be available to developers on OpenRouter, Fireworks, and Baseten, with weight tuning for the first time. Microsoft also describes Frontier Tuning, using reinforcement learning environments to let MAI models learn from an organization’s workflows, and a collaboration with Mayo Clinic to co-create a frontier AI model for healthcare.