Microsoft AI internal model outperforms OpenAI equivalent using quarter compute

Microsoft AI Chief Executive Mustafa Suleyman told employees at a town hall meeting on October 15, 2026, that the company built an in-house artificial intelligence model that outperforms an OpenAI base model. Business Insider reported that the internal model beat an OpenAI equivalent on specific pre-training metrics. [1]

According to Business Insider, Suleyman stated during the meeting that Microsoft AI developed a pre-trained base model that is stronger than the model powering OpenAI's GPT 5.6 Sol. He added that Microsoft AI achieved these performance results while using only one-quarter of the computing power that OpenAI utilized to train its model. [1]

Pre-training represents the initial phase of AI model development, during which a system learns underlying patterns from massive volumes of images, text, and other data inputs. Business Insider reported that this foundational training stage traditionally requires significant quantities of costly computing infrastructure. [1]

The performance claims highlight ongoing internal development at Microsoft AI, where Suleyman, Jordan Hoffmann, and former Inflection staff were hired to build language models. OpenAI, which is led by Chief Executive Sam Altman, released GPT-6.1 Sol in 2026 and counts Microsoft as one of its primary partners. [6][7][8]