Key facts
- Users report that OpenAI's GPT-6 Astra model has become "dumber" and less capable since its launch.
- Complaints suggest a decrease in writing quality and reasoning ability, with some users comparing it to previous model regressions.
- Some users have switched back to older models like GPT-5.6 Sol due to perceived performance drops in Astra.
- Speculation among users suggests OpenAI may have reduced the "juice value" or computing effort allocated to Astra to cut costs.
- OpenAI executive Tibo Sottiaux previously confirmed experimentation with reasoning effort settings on GPT-5.6 Sol.
- GPT-6 Astra is OpenAI's first model capable of identifying and chaining unknown software vulnerabilities without human guidance.
Users of OpenAI's latest AI model, GPT-6 Astra, are reporting a significant decline in its performance and writing quality just a week after its highly anticipated launch. Complaints flooding social media platform X suggest the model has become "dumber," mirroring a similar backlash experienced with OpenAI's previous flagship model, GPT-5.6 Sol, in July.
Users like the pseudonymous developer synthwavedd have expressed disappointment, with one posting, "Astra feels significantly dumber for me today." Developer Pranjal Paliwal, who had initially praised Astra, later reviewed the code it generated and concluded, "We don't have AGI. We have a regression." The term AGI, or artificial general intelligence, describes a machine capable of human-level cognitive tasks, a concept OpenAI's president invoked at Astra's launch.
The complaints often center on a perceived drop in quality despite faster response times. Users speculate that OpenAI may have reduced the "juice value"—a shorthand for the computing effort the model expends—to cut operational costs. This suspicion is fueled by past instances, such as when OpenAI executive Tibo Sottiaux confirmed experimentation with reasoning effort settings on GPT-5.6 Sol.
Empirical tests by users like Salio and researcher Md Ismail Sojal, using identical prompts on launch-day Astra versus the current version, reportedly showed worse results from the latter. Some users, including Dax Raad, have reverted to using GPT-5.6 Sol, citing increased costs for diminished returns with Astra. Mustafa Sahinli compared Astra's current performance unfavorably to Anthropic's Claude Opus 4.6 after its own post-launch criticism.
However, not all users agree that Astra has been deliberately nerfed. The user Antikythera argued that the model's performance hasn't changed and that users are now simply noticing its flaws after the initial hype subsided. Theo, founder of T3Chat, suggested Astra is more inconsistent, capable of producing both outstanding and poor results, with users now more frequently posting the negative outcomes as the "honeymoon" period ends.
OpenAI has not yet issued a statement regarding the Astra complaints, unlike the response to the GPT-5.6 Sol backlash. Astra is noted as OpenAI's first model capable of identifying and chaining unknown software vulnerabilities without human guidance. Despite the performance concerns, GPT-6 Astra is priced at $10 per million input tokens and $50 per million output tokens, 2.5 times the cost of GPT-5.6 Sol at its launch.