Users Report Performance Regression in OpenAI's GPT-6 Astra After Launch

Users on X have posted complaints that OpenAI's latest flagship model, GPT-6 Astra, appears to have become less capable one week after launch. According to ChainCatcher, developers and researchers said the model now gives faster but lower-quality answers, with some comparing current results against those from release day and finding a decline. Anonymous developer synthwavedd said Astra felt noticeably worse and joked about a post-launch "lobotomy." Pranjal Paliwal, who had previously praised the model, said after reviewing code generated by Astra that "we do not have AGI, we have a performance regression." Pankaj Kumar said the symptoms included faster responses and poorer quality, and suggested OpenAI may have reduced the model's "juice value," referring to the compute used before answering. Salio and researcher Md Ismail Sojal used the same prompts to compare Astra on launch day and at present, and both obtained worse results. Theo, founder of T3Chat, said Astra may simply be less stable than Claude Fable and that users began sharing more negative examples after the initial honeymoon period ended. Some users also pointed to a similar pattern with OpenAI's previous flagship, GPT-5.6 Sol, in July. OpenAI executive Tibo Sottiaux denied intentionally weakening the model at the time, but said the company had been experimenting with "reasoning effort" settings. OpenAI has not issued a similar statement about Astra.