Microsoft announces new AI models to reduce dependence on OpenAI and reduce costs

microsoft is a major player in the artificial intelligence boom, providing key cloud infrastructure and services and acquiring multibillion-dollar stakes in OpenAI and Anthropic. Currently, the company is making all-out efforts to compete with its own models.

At the Build developer conference in San Francisco on Tuesday, Microsoft announced MAI-Code-1-Flash, the first model to take written instructions from people and spit out source code for applications and websites. The AI ​​coding market (vibe coding) has recently been booming with developers and people without a technical background creating sophisticated software using text-based prompts.

For Microsoft, using the primary model would drive up costs, so it makes economic sense to offer a proprietary model that can be passed on to developers. Microsoft can run models on its own Azure cloud infrastructure and avoid paying third parties like OpenAI. In May, google announced a Gemini 3.5 flash model that allows you to write code and perform other tasks and run it in the search company’s own data center.

In addition to MAI-Code-1-Flash, Microsoft has introduced an inference model, MAI-Thinking-1, to increase the efficiency of both products.

The inference model is medium-sized and “built for high efficiency and performance, but importantly, low token costs,” Kyle Daigle, Microsoft’s head of developer marketing and head of GitHub operations, said in a blog post. Tokens are the building blocks of data that models read, process, and produce, and their use determines developer costs.

Microsoft is looking to play at more layers of the AI ​​stack as OpenAI and Anthropic continue to record historic growth and push towards the public market. Anthropic secretly filed for an IPO on June 1, and OpenAI is also aiming to go public, possibly later this year. Microsoft has invested $13 billion in OpenAI and $5 billion in Anthropic, making its models available through Azure.

MAI-Thinking-1 is available in private preview through Microsoft Foundry, a service for integrating models into applications. Customers can express interest in testing a model before it is widely available.

Customers can improve the accuracy of their inference models by incorporating their own data.

“What we’ve just seen is a pretty significant change,” Microsoft CEO Satya Nadella said on stage. “We believe the time has come for all companies to move from consuming the frontier model to fully participating in the frontier of the frontier ecosystem.

After refining the model to meet consulting firm McKinsey’s needs, Microsoft was able to outperform OpenAI’s GPT 5-5 and be 10x more cost-effective, Microsoft AI CEO Mustafa Suleiman said.

Daigle says this coding model is “very efficient for inference” and is available in the GitHub Copilot AI coding service and the Visual Studio Code text editor.

Also on Tuesday, Microsoft will announce its latest cloud-based models for speech recognition, synthetic speech generation, and image generation, as well as a smaller Aion model that can run on Windows PCs.

Watch: Charles Lamanna, Microsoft EVP Copilot, Agents and Platforms: Learn more about Fort Knox