Microsoft AI chief says the company beat OpenAI at its own game, using a quarter of the computing power

Microsoft AI CEO Mustafa Suleyman told workforce an in-house example was improved than a akin offering from OpenAI, among the application giant's biggest customers and partners.

"We now have a stronger pre-trained basis example than the example that currently powers GPT 5.6 Sol," Suleyman stated during a town hall gathering alongside workforce on Thursday. "And in fact, we used one-quarter of the compute that OpenAI used to train their model."

Pre-training is the archetypal phase of construction an AI model, whenever it learns patterns from huge amounts of text, images, and another information. This takes enormous amounts of costly computing power, so Microsoft could preserve prosperity if it's worked out a way to do this additional efficiently.

Microsoft announced the model, MAI-Base-2, earlier on Thursday.

Suleyman stated Microsoft privately tested this example against OpenAI models, including one nicknamed "Spud" that powers GPT-5.6. He added that Microsoft's example performed improved on tests measuring how fine AI can comprehend and foretell patterns in information.

The comparisons are according to specialized data Microsoft has concerning OpenAI's models, Suleyman said.

"Privately and confidentially, we've additionally compared it to the OpenAI models," Suleyman told employees, during suggesting that Microsoft has particular admission to data on how OpenAI models perform.

It's notable that Suleyman would choose to emphasize restrictions of OpenAI models, equal in personal discussions. Microsoft has invested billions of dollars in OpenAI, and owns a interest in the startup that is apt value additional than $100 billion. Any criticism of OpenAI could possibly dent the value of this huge investment.

The two companies additionally have a important partnership, alongside Microsoft offering OpenAI models to its Azure haze customers. Microsoft has used OpenAI models for its own products specified as Copilot, too.

But the association between the two companies has frayed in the former twelvemonth or so. Increasingly, Microsoft has tried to branch out from OpenAI, including by operating on its own models.

Suleyman additionally stated Microsoft's new AI example used "12 times small computation than was used for DeepSeek, for example, on compute efficiency."

Suleyman didn't define which DeepSeek model, but a public blog post described the pre-training of MAI-Base-2 being a lot additional productive than a akin offering from the Chinese lab, called DeepSeek-V4-Pro-Base.

Microsoft in March created a new blended squad for its business and person Copilot products and reassigned Suleyman to concentration on construction AI models in the company's superintelligence unit.

Microsoft declined to comment and OpenAI did not react to a petition for comment on Thursday.

Have a tip? Contact this newsman via email at [email protected] or Signal at +1-425-344-8242. Use a individual email location and a nonwork device; here's our guide to sharing data securely.