Microsoft Launches Its Own AI Models at Build Conference, Aiming to Reduce Dependence on OpenAI

Microsoft on Tuesday introduced a pair of proprietary artificial intelligence models at its annual Build developer conference in San Francisco, marking the company's most deliberate push yet to compete at the model layer of the AI stack rather than simply distributing the work of others.

The two models are MAI-Code-1-Flash, Microsoft's first entry into AI-assisted software development, and MAI-Thinking-1, a reasoning model aimed at tasks requiring multi-step logic. Both are positioned around cost efficiency — a pointed contrast with the increasingly expensive frontier models offered by OpenAI and Anthropic.

MAI-Code-1-Flash takes written descriptions and generates source code for applications and websites. It is already available inside GitHub Copilot and the Visual Studio Code text editor.

MAI-Thinking-1, described as medium-sized and "built for high efficiency and performance, but importantly, at a low-token cost," is currently in private preview through Microsoft Foundry, the company's service for integrating models into applications. Customers can register interest before broad availability, and the company said they will be able to improve the model's accuracy by incorporating their own proprietary data.

The economic logic behind the push is direct: by running its own models on Azure infrastructure, Microsoft can avoid paying licensing fees to third-party providers such as OpenAI, and pass at least some of those savings to developers.

Microsoft CEO Satya Nadella framed the moment as a strategic inflection point. "What you just saw is a pretty significant shift," Nadella said onstage. "We believe the time has come for every company to just move from consuming a frontier model to fully participating at the frontier in the frontier ecosystem."

Mustafa Suleiman, CEO of Microsoft AI, offered a more pointed benchmark: after tuning its models for the needs of consulting firm McKinsey, Microsoft was able to outperform OpenAI's GPT-5.5 with 10 times better cost efficiency, he said.

Kyle Daigle, Microsoft's developer marketing chief and GitHub's operating chief, wrote in a blog post that the coding model is "inference ultra-efficient," underscoring the company's emphasis on running inference at lower compute cost.

The announcements come as Microsoft finds itself in an unusual position. The company has invested $13 billion in OpenAI and $5 billion in Anthropic, making their models available through Azure while also building its own competing offerings. Both OpenAI and Anthropic are now moving toward public markets — Anthropic said Monday that it had confidentially filed for an IPO, while OpenAI is also pursuing an offering potentially this year.

Google introduced its own cost-focused Gemini 3.5 Flash model in May, capable of coding and other tasks, signaling that efficiency-oriented models have become a competitive front across the industry.

Beyond the two headline models, Microsoft also revealed updated cloud-based tools for speech recognition, synthetic voice generation, and image generation, along with small Aion models designed to run locally on Windows PCs.

The Build announcements reinforce that Microsoft is no longer content to sit primarily as a cloud host and strategic investor. As OpenAI and Anthropic grow toward the public markets and capture an expanding share of developer mindshare, Microsoft's decision to field its own proprietary models signals a deepening competition that runs directly through its most important AI partnerships.

Disclaimer