What researchers are saying

Altman says the new models still need work to be pleasant to use

Sam Altman, chief executive of OpenAI

Altman wrote that GPT-6.2 Astra and Sol are highly capable and efficient, but acknowledged that they still need work on being nice to work with.

Sam Altman, chief executive of OpenAI, wrote on X on the day of the release that GPT-6.2 Astra and Sol are highly capable and efficient models. In the same post he acknowledged that they still need work on being nice to work with, separating raw capability from the experience of using the model.

The remark came as the first reviews from heavy users were negative. One widely shared review described Astra as frustrating to use, with no sense of the limits of its own knowledge, outputs that are hard to parse and a tendency to run in circles on long tasks. Altman gave no timetable for improvements.

The comment points at a measure benchmarks do not capture. As frontier models converge on headline scores, how easily people can work with them is becoming a separate axis of competition.