マイクロソフトAI CEO、低計算量の新モデルがOpenAIを上回ると発言

Microsoft is scheduled to release its proprietary model, MAI-Base-2, on October 15. On the same day, Microsoft AI CEO Mustafa Suleyman stated at an employee meeting that internal testing showed the model outperformed the base model currently used for GPT-5.6 Sol in benchmarks evaluating the understanding and prediction of information patterns. He also claimed that the pre-training compute used by Microsoft was only one-quarter of that used for the corresponding OpenAI model. These comparisons with OpenAI stem from Suleyman's internal remarks, as Microsoft's publicly released technical articles did not provide complete comparative data available for external verification. According to Microsoft's disclosure, MAI-Base-2 is a Mixture-of-Experts model with 2.8 trillion total parameters and 80 billion active parameters. The company stated that in its evaluations, the model's pre-training computational efficiency exceeded that of DeepSeek-V4-Pro-Base. However, MAI-Base-2 is currently undergoing post-training, and the test performance of the base model cannot be directly equated with the final product experience for end-users. While maintaining investment and product partnerships with OpenAI, Microsoft is increasing its investment in proprietary models. The release of MAI-Base-2 indicates that Microsoft is attempting to control more stages of the pipeline from model training to product application, but its cost and performance advantages still need to be verified through subsequent public testing.