OpenAI GPT-6.2 Sol regresses deep research
The author argues that OpenAI's GPT-6.2 Sol is a regression for deep research compared to its predecessor, GPT-6.1 Sol, because it writes reports that are half as long with fewer details and citations. Despite being slightly cheaper and faster, turning up the reasoning effort on the new model did not improve report depth or length. Consequently, the author reverted their application back to GPT-6.1 Sol. Open article →