MICROSOFT BEAT OPENAI AT ITS OWN GAME — WITH A QUARTER OF THE COMPUTE

Microsoft AI chief Mustafa Suleyman told employees at a town hall that the company has built a pre-trained model stronger than the one behind OpenAI's GPT-5.6 Spur while using only a quarter of the computing resources, Business Insider reported. Microsoft unveiled the model, MAI-Base-2, on October 15. Suleyman said Microsoft privately compared it with several OpenAI models, using technical information it holds on those models, and that MAI-Base-2 scored better on tests of understanding information and predicting patterns.

Suleyman also cited a comparison with DeepSeek, saying Microsoft's approach used 12 times less compute, without naming the DeepSeek model. The claim underscores a shifting relationship: Microsoft has invested more than $10 billion in OpenAI and built products such as Copilot on its models, but is now reducing that reliance by developing its own models and working with Anthropic.