Microsoft AI CEO says new model beats OpenAI on pre-training metrics using less compute

Microsoft AI CEO Mustafa Suleyman told employees the company’s in-house MAI-Base-2 model was stronger than a comparable OpenAI base model on certain pre-training metrics. He said Microsoft used one-quarter of the compute OpenAI used to train its model. The key takeaway for readers: Microsoft is presenting AI model efficiency as a strategic advantage, not just raw performance.

Pre-training is the early stage of model development, when an AI system learns patterns from large amounts of text, images, and other information. Because that process requires expensive computing power, using less compute could help reduce costs if the performance claims hold up. Suleyman said Microsoft’s model performed better on tests measuring how well AI understands and predicts patterns in information.

The comments stand out because Microsoft is both a major OpenAI investor and a key partner, offering OpenAI models to Azure customers and using them in products such as Copilot. Business Insider reported that the relationship has frayed in the past year or so as Microsoft increasingly works on its own models. Suleyman also said Microsoft privately compared MAI-Base-2 with OpenAI models using technical information Microsoft has about them.

Suleyman also said Microsoft’s new AI model used “12 times less computation” than DeepSeek, though he did not specify which DeepSeek model in that comment. Business Insider noted that a Microsoft public blog post described MAI-Base-2 pre-training as more efficient than DeepSeek-V4-Pro-Base. Microsoft declined to comment, and OpenAI did not respond to a request for comment on Thursday.