Microsoft has developed an AI model it claims is more powerful than OpenAI’s using a quarter of the computing resources, Business Insider reported on Oct. 15.

Mustafa Suleyman (무스타파 술레이만), CEO of Microsoft AI, told a staff town hall meeting that the company had secured a pre-trained model more powerful than the one running GPT 5.6 Spur. He stressed it used only a quarter of the computing resources OpenAI used to train its model.

Microsoft unveiled MAI-Base-2 on Oct. 15. Pre-training is the first stage in which an AI model learns patterns from vast amounts of text and images. Improving efficiency can help Microsoft save costs because it requires massive computing resources.

Suleyman said Microsoft privately compared MAI-Base-2 with several OpenAI models. The comparisons included a GPT-5.6-based model referred to internally at OpenAI as Spur.

In tests measuring the ability to understand information and predict patterns, Microsoft’s model showed better results, Suleyman said. He said the comparisons were conducted based on technical information on OpenAI models held by Microsoft.

He also released results from a comparison with DeepSeek. Suleyman said it used 12 times less compute resources than DeepSeek in terms of computing efficiency, but did not disclose which model.

It is a notable move for Microsoft executives to mention limitations of OpenAI models, Business Insider said.

Microsoft has invested more than $10 billion in OpenAI, and the value of its stake is expected to exceed $100 billion. Microsoft has provided OpenAI models to Azure cloud customers and used them in its own products such as Copilot, but the relationship between the two companies has changed markedly over the past year or so.

Microsoft is reducing its reliance on OpenAI by developing its own models and cooperating with Anthropic.