Within a week of its launch, GPT-6 Astra is facing widespread user complaints about diminished performance. Developers posting on social media report that the model's output quality has declined, with some describing answers as faster but less accurate than initial release-day results.
Several users documented the shift by running identical prompts against launch-day and current versions of the model, with researchers Salio and Md Ismail Sojal reporting noticeably worse results from the current version using the same settings. Developer Pranjal Paliwal, who initially praised the model, reversed his assessment after reviewing code samples, stating the model exhibits both exceptional and surprisingly poor performance.
Speculation centers on whether OpenAI reduced the model's "reasoning effort"—the computing resources allocated to processing before generating answers. OpenAI executive Tibo Sottiaux previously confirmed the company had been experimenting with reasoning effort settings for GPT-5.6 Sol, though denied deliberately weakening that model when similar complaints emerged in July.
Not all observers believe intentional changes occurred. Some argue the model performed similarly at launch but users are now noticing flaws after the initial enthusiasm has faded. T3Chat founder Theo suggested Astra produces inconsistent results, with some outputs exceptional and others poor, and that heightened attention to failures may reflect shifts in user expectations rather than actual model degradation.
GPT-6 Astra remains OpenAI's first model to cross what the company calls the critical cybersecurity risk threshold, capable of identifying and chaining together unknown software vulnerabilities without human guidance. The model is priced at $10 per million input tokens and $50 per million output tokens, 2.5 times higher than the launch pricing of its predecessor.
OpenAI has not issued a public statement addressing the performance concerns, unlike its response to similar complaints about GPT-5.6 Sol.
