Microsoft developing AI reasoning models to compete with OpenAI, The Information reports

Microsoft is reportedly developing its own artificial intelligence reasoning models in an effort to compete with OpenAI, with potential plans to make them available to developers. According to a report from The Information on Friday, the company has already begun testing alternative models from xAI, Meta, and DeepSeek as possible replacements for OpenAI's technology in its Copilot assistant.

Despite being a major OpenAI backer, Microsoft is looking to reduce its reliance on the company, even though their early partnership gave it a significant advantage in the AI race among major tech firms. Back in December, Reuters reported that Microsoft was actively working on integrating both internal and third-party AI models into its Microsoft 365 Copilot to diversify its AI technology and cut costs.

Originally, Microsoft 365 Copilot, introduced in 2023, was built around OpenAI’s GPT-4 model. However, Microsoft's AI division, led by Mustafa Suleyman, has since developed and trained a family of AI models internally known as MAI. These models are reported to perform nearly as well as OpenAI and Anthropic’s leading models on industry-standard benchmarks.

Additionally, the team is focusing on training reasoning models utilizing "chain-of-thought" techniques, which enhance intermediate reasoning abilities when handling complex problems. According to The Information, Microsoft is already testing its MAI models—larger than its previous Phi models—as potential replacements for OpenAI’s models in Copilot.

There is also consideration within Microsoft to release the MAI models as an API later this year, allowing developers to integrate them into their own applications. Microsoft and OpenAI have not yet commented on these reports.