Microsoft Corp. is testing artificial intelligence (AI) models to rival those of OpenAI that it could sell to developers in a move that could portend increased competition between the two allies.
The software giant, which has pumped more than $13 billion into OpenAI, is developing models from OpenAI, Chinese AI startup DeepSeek, Meta Platforms Inc. and Elon Musk’s xAI as potential OpenAI replacements in Microsoft Copilot, according to a report in The Information, which cited a person involved in the project.
Microsoft’s AI division, under the helm of Mustafa Suleyman, has completed training for a family of models, internally called MAI, that perform on par with leading models from Anthropic and OpenAI on generally accepted benchmarks. The division is also training reasoning models which use chain-of-thought techniques that could compete directly with OpenAI’s models, the report said.
Suleyman’s team is already experimenting with swapping out MAI models for OpenAI’s models in Copilot, the report said, and is considering releasing the MAI models later this year as an application programming interface. Such an arrangement would let third-party developers integrate MAI models into their own apps.
Microsoft and OpenAI were not immediately available for comment.
The machinations by Microsoft represent a move by the company to increasingly be less dependent on OpenAI’s ChatGPT as the two companies make the inevitable journey from cooperation to competition. For a few years, Microsoft has poured billions of dollars into the Sam Altman-led startup to become a leader in AI while also supporting its technology as a bedrock of Microsoft’s strategy. When Microsoft announced Copilot in 2023, for example, it said it used OpenAI’s ChatGPT-4 model.
But there have been growing signs that Microsoft is weaning itself from OpenAI. In December, Reuters reported that the Redmond, Wash.-based behemoth had been working on adding internal and third-party AI models to undergird Microsoft 365 Copilot and rely less on underlying technology from OpenAI.
Microsoft’s embrace of reasoning models underscores a general movement by AI’s biggest players to embrace the new category of specialized language models. Reasoning models are designed to break down complex problems into smaller, manageable steps and solve them through explicit logical reasoning. Unlike general-purpose large language models (LLMs) that generate direct answers, reasoning models are specifically trained to show their work and follow a more structured thought processes.
