Microsoft’s AI head, Mustafa Suleyman, told employees the company’s new base model, MAI-Base-2, outperformed an OpenAI system said to power GPT-5.6, while requiring about a quarter of the training compute. Suleyman said the conclusions were drawn from internal benchmarks and technical information Microsoft has about OpenAI’s models, and added that MAI-Base-2 trained with far less compute than a comparable DeepSeek model. The assertions highlight Microsoft’s push to develop its own foundation models even as it remains a major OpenAI investor and reseller—an evolving dynamic with potential cost and strategy implications for Azure and Copilot. Microsoft declined to comment; OpenAI did not respond to a request for comment.
Microsoft AI chief claims in-house model outperforms OpenAI’s while using just 25% of the compute
- Trending
- Comments
- Latest
