Microsoft on June 2 (local time) unveiled its self-developed AI coding model, MAI-Code-1-Flash, at its annual Build developer conference held in San Francisco.

According to a CNBC report, MAI-Code-1-Flash generates application and website source code when users enter a text description. It can be used in GitHub Copilot and Visual Studio Code.

Microsoft also released its reasoning model MAI-Thinking-1. The company said MAI-Thinking-1 is a mid-sized model and its strength is that token costs are cheaper than others.

MAI-Thinking-1 will be provided as a private preview through Microsoft Foundry.

Mustafa Suleyman (무스타파 술레이먼), CEO of Microsoft AI, stressed that optimising the model to meet the needs of consulting firm McKinsey achieved performance that surpasses OpenAI GPT-5.5 while being 10 times more cost-efficient.

Satya Nadella (사티아 나델라), Microsoft CEO, said, "As every company is at a stage of consuming frontier models, it is time to shift to directly participating in the frontier ecosystem."

Microsoft has provided the models of the two companies through Azure after investing $13 billion in OpenAI and $5 billion in Anthropic. Launching its own models is a strategy to reduce reliance on external vendors' frontier models and offer developers cheaper alternatives.

Microsoft also introduced cloud models for speech recognition, speech generation and image generation, as well as small Aion models that run on Windows PCs.