A week ago, GPT-6 Astra stunned users by rebuilding Manhattan street by street inside a game engine. Today, those same users are flooding social media with screenshots to ask OpenAI what happened.
Popular pseudonymous developer synthwavedd posted on X that Astra feels significantly dumber. They called the drop in performance the post-launch lobotomy.
Developer Pranjal Paliwal initially praised the model before actually reading the code Astra wrote for him. After inspecting the output, Paliwal took his words back.
OpenAI president used the term AGI at Astra's launch to describe cognitive parity with humans. A week later, users are applying the label in reverse.
Other developers are listing similar symptoms. Pankaj Kumar noted faster answers, worse quality, and a suspicion that OpenAI reduced the juice value.
Juice value is informal shorthand for the computing effort a model spends thinking before responding. Users suspect OpenAI turned down that dial after launch demos concluded.
Some developers tested the theory directly. Salio and researcher Md Ismail Sojal ran identical prompts against launch-day Astra and the current version. Both received worse results from the modern iteration.
Dax Raad builds the coding tool Opencode. His team switched back to Astra's predecessor, GPT-5.6 Sol, after finding the new spend doubled for unworthy downsides.
Not everyone agrees OpenAI intentionally nerfed the model. Pseudonymous user Antikythera argued that users simply stopped being dazzled long enough to notice the model's existing flaws.
T3Chat founder Theo suggested Astra is just highly inconsistent. Theo noted the model produces both outstanding and profoundly stupid code because the honeymoon phase has ended.
OpenAI experienced the exact same cycle in July with GPT-5.6 Sol. Users reported its reasoning mode went shallow overnight, prompting a denial of intentional weakening from executive Tibo Sottiaux.
OpenAI has not yet issued a statement regarding the Astra complaints. The model still costs $10 per million input tokens and $50 per million output tokens.
