Microsoft unveils new AI models lessen reliance on OpenAI, lower costs

Microsoft It has been a major player in the AI ​​boom, providing key cloud infrastructure and services and taking multibillion-dollar equity stakes in OpenAI and Anthropic. Now the company is making a concerted effort to compete with proprietary models.

At the Build developer conference in San Francisco on Tuesday, Microsoft announced MAI-Code-1, its boot model that takes written comments from people and distributes source code for apps and websites. The AI ​​coding market, or jitter coding, has taken off recently as developers and people without technical backgrounds use text-based prompts to produce complex software.

For Microsoft, providing its own models has economic benefits that can be passed on to developers as costs increase in the use of leading models. Microsoft can run its models on its own Azure cloud infrastructure and avoid paying third parties like OpenAI. In May, Google announced the Gemini 3.5 Flash model, which runs in the search company’s data centers, can code and perform other tasks.

In addition to MAI-Code, Microsoft is introducing a reasoning model, MAI-Thinking, improving the efficiency of both offerings.

The reasoning model is mid-sized and “built for high efficiency and performance, but more importantly, at low cost,” Kyle Daigle, Microsoft’s head of developer marketing and GitHub business chief, wrote in a blog post. Tokens are used by developers to pay for model usage.

As OpenAI and Anthropic continue to post historic growth and move toward the public market, Microsoft is looking to play into more layers of the AI ​​stack. Anthropic said Monday that it has secretly filed for an IPO, and that OpenAI is also potentially pursuing a bid this year. Microsoft invested $13 billion in OpenAI and $5 billion in Anthropic while making its models available on Azure.

MAI-Thinking-1 is available as a private preview through Microsoft Foundry, a service for integrating models into applications. Customers can: show interest Testing the model before widespread availability.

Daigle wrote that the coding model is “extremely efficient inference” and is available in the GitHub Copilot AI coding service and the Visual Studio Code text editor.

Also on Tuesday, Microsoft is unveiling updated cloud-based models for speech recognition, synthetic voice rendering and rendering, as well as smaller Aion models that can run on Windows PCs.

WRISTWATCH: Charles Lamanna, Microsoft EVP Copilot, Agents and Platform: A Deep Look at Fort Knox