MICROSOFT AI CEO SAYS NEW MODEL BEATS OPENAI USING LESS COMPUTE

What happened: Microsoft AI CEO Mustafa Suleyman told employees that Microsoft's in-house model, MAI-Base-2, is a stronger pre-trained base model than the one powering GPT 5.6 Sol, and that it was trained with one-quarter of the compute OpenAI used. Key numbers: One-quarter of OpenAI's compute for training. 12 times less computation than DeepSeek. Why it matters: Suleyman's private criticism of OpenAI models could dent the value of Microsoft's stake, and it reflects a relationship that has frayed as Microsoft works on its own models.

Microsoft AI CEO Mustafa Suleyman told employees that Microsoft's in-house model, MAI-Base-2, is a stronger pre-trained base model than the one powering GPT 5.6 Sol, and that it was trained with one-quarter of the compute OpenAI used.

Suleyman's private criticism of OpenAI models could dent the value of Microsoft's stake, and it reflects a relationship that has frayed as Microsoft works on its own models.