Sam Altman: GPT 5.5 an average math professor. 5.6 top one or two percentile. Astra a little bit better. Internal model can do things that the best mathematicians in the world cannot.

Sam Altman stated that GPT 5.5 performed at the level of an average math professor, while GPT 5.6 achieved the proficiency of a top one or two percentile math professor. He also mentioned that Astra is slightly better, and an internal model possesses capabilities that even the world's best mathematicians cannot match.

This report offers a rare, specific comparison of GPT 5.5, GPT 5.6, and Astra's mathematical capabilities, unlike general statements about AI progress.