Microsoft AI has launched seven new in-house MAI models in the areas of reasoning, coding, image generation, voice and transcription, as well as a healthcare AI collaboration with the Mayo Clinic. This announcement, made on June 2, 2026, introduces seven new models: MAI-Thinking-1, MAI-Code-1-Flash, MAI-Image-2.5, MAI-Image-2.5 Flash, MAI Transcribe-1.5, MAI-Voice-2, and MAI-Voice-2-Flash.
Microsoft AI says these models are part of a multifaceted family designed to work on real-world tasks and be integrated into products used by individuals, developers, and organizations. The launch also introduces Microsoft Frontier Tuning, a way to customize models to organizations’ specific workflows. For developers and education providers focused on AI skills, this launch brings new model capabilities in coding, software engineering, voice, transcription, and enterprise workflow training. MAI-Code-1-Flash is built for GitHub Copilot, Visual Studio Code, and Microsoft’s broader stack.
Seven new models join the MAI family
MAI-Thinking-1 is Microsoft AI’s flagship reasoning model. Microsoft AI states that the model has been trained from the ground up on pure data, with no data from third-party models, and is comparable to leading models on key software engineering benchmarks. Microsoft AI CEO Mustafa Suleyman said MAI-Thinking-1 is a mixture-of-experts model with 35 billion active parameters and a 256,000-token context window.
MAI-Code-1-Flash is a coding AI model designed to work efficiently and help with programming in tools like GitHub Copilot, Visual Studio Code, and other Microsoft platforms. Suleiman explained that the model has five billion parameters and is “specifically optimized for VS Code and the GitHub Copilot CLI.”
Microsoft said MAI-Image-2.5 and MAI-Image-2.5 Flash support text-to-image generation and image editing capabilities. MAI Transcribe-1.5 is designed for transcription in 43 languages, while MAI-Voice-2 offers speech generation in 15 languages and can customize voice from a small sample, with safeguards to prevent misuse.
Frontier Tuning will bring organization-centric AI models
Microsoft AI describes these environments as training spaces where models can learn from the sequences of tasks, decisions, and actions within an organization. The company says that customer data remains within the organization’s environment and control.
Microsoft AI says initial tuning work has yielded improvements in efficiency and performance. Microsoft said a Frontier-Tuned model for Excel achieved performance comparable to GPT-5.4 while operating with significantly greater efficiency.
The model tuned for McKinsey achieved the highest win rate of all models tested and cost approximately 10 times less.
Mayo Clinic Partnership to Develop AI Model for Healthcare
Microsoft and Mayo Clinic are collaborating to develop a healthcare AI model designed for clinical reasoning and healthcare applications.
Microsoft AI will test the new technology inside Mayo Clinic first. After it is tested and approved, the model will be available for other organizations through Azure Foundry. This launch also further expands Microsoft AI’s broader strategy. Microsoft AI says it is building a SuperIntelligence Lab, which includes in-house models trained from scratch, clean and properly licensed datasets, technical and security reports, and co-design work with Maia 200 silicon.
Microsoft AI says it will rapidly expand computing and capabilities over the next year. MAI models are available now, MAI-Voice-2-Flash will be available soon, and Mayo Clinic’s healthcare model will debut in Mayo Clinic’s own environment before broader availability through Azure Foundry.
