Microsoft Launches Proprietary AI Models to Cut OpenAI Dependency

CNBC reported Tuesday that Microsoft unveiled a suite of proprietary Microsoft AI models at its annual Build developer conference in San Francisco. The move signals an aggressive push into a space long dominated by OpenAI, Anthropic, and Google.

Microsoft Enters the AI Coding Race

The headline announcement was MAI-Code-1-Flash, Microsoft’s first original model purpose-built for code generation. Developers and non-technical users can describe what they want in plain language. The model then produces working application or website code. This market, informally known as “vibe coding,” has expanded rapidly over the past year.

Alongside the coding model, Microsoft introduced MAI-Thinking-1, a mid-sized reasoning model. According to a blog post from Kyle Daigle, Microsoft’s developer marketing chief and GitHub’s operating chief, the reasoning model is engineered for high efficiency at low token costs. Tokens are the currency developers use to pay for AI model consumption.

Also Read: What Is Vibe Coding and Why Are Developers Obsessed With It

Why Microsoft Is Building Its Own Stack

Until now, Microsoft’s primary role in the AI boom has been infrastructure and investment. The company has committed $13 billion to OpenAI and $5 billion to Anthropic, while routing both firms’ models through its Azure cloud platform. That arrangement carries real costs. Running proprietary models on Azure allows Microsoft to bypass third-party fees entirely, with potential savings passed to developers.

CEO Satya Nadella told the Build audience that the moment has arrived for companies to move beyond simply consuming frontier AI models. He framed Microsoft’s shift as full participation in the frontier ecosystem rather than passive consumption.

Benchmarks and the Competitive Backdrop

Microsoft AI CEO Mustafa Suleiman offered a pointed benchmark claim. After fine-tuning MAI-Thinking-1 for consulting firm McKinsey, the model reportedly outperformed OpenAI’s GPT-4.5 while delivering ten times better cost efficiency. That framing lands as both rivals prepare for public listings. Anthropic confirmed Monday it had confidentially filed for an IPO, and OpenAI is pursuing a potential offering later this year.

MAI-Thinking-1 is currently available through Microsoft Foundry in private preview. Developers can register interest ahead of a broader rollout, and the model supports fine-tuning with proprietary data to improve accuracy.

MAI-Code-1-Flash is already live inside GitHub Copilot and Visual Studio Code. Microsoft also announced refreshed cloud models covering speech recognition, voice synthesis, and image generation, plus lightweight Aion models designed to run locally on Windows PCs.

Read Next: Anthropic Confidentially Files for IPO as AI Rivals Race Toward Public Markets