Microsoft AI chief says the company beat OpenAI at its own game, using a quarter of the computing power (www.businessinsider.com)

In a recent announcement during a town hall meeting, Microsoft AI chief Mustafa Suleyman claimed that the company has developed a pre-trained base model named MAI-Base-2 that surpasses OpenAI’s GPT-5.6 "Spud" in performance, using just a quarter of the computational resources. This significant achievement highlights Microsoft's advancements in model efficiency, showcasing their ability to pre-train a robust AI model while dramatically reducing computational costs. Pre-training is a critical stage in AI development, where models learn from extensive datasets, making this efficiency breakthrough potentially game-changing for the industry's resource utilization. Suleyman's comments come amidst a complicated relationship between Microsoft and OpenAI, with Microsoft being a major investor in the latter. His remarks subtly critique the limitations of OpenAI’s models, despite their ongoing partnership. Additionally, Microsoft claims that MAI-Base-2 is significantly more computationally efficient, utilizing 12 times less computation than a similar model from the Chinese lab DeepSeek. This development underscores Microsoft's commitment to innovation in the AI space as it aims to establish its own identity in a rapidly evolving landscape.