Just a week ago, GPT-6 Astra was impressing users by rebuilding Manhattan street by street inside a game engine. This week, many of those same users are posting screenshots on X and asking OpenAI what happened to the model.
The pseudonymous developer synthwavedd wrote, 「Astra feels significantly dumber for me today. Was only a matter of time before The Post-Launch Lobotomy. Shame.」
The complaint follows a pattern familiar to heavy AI users, who often say their preferred chatbots or agents have been “nerfed.” In many cases, there are reasonable explanations for perceived drops in quality. But when many people begin making the same complaint at the same time, the issue draws more scrutiny. Right now, social media is filled with examples of users turning on GPT-6 Astra with unusual speed.
Developers describe faster answers and weaker output
Developer Pranjal Paliwal, who had previously praised Astra, went back and reviewed code the model wrote for him. He did not like what he saw.
He posted, 「We don't have AGI. We have a regression.」
AGI, short for artificial general intelligence, is the term commonly used for a machine that can do essentially any cognitive task a human can do. OpenAI’s president used that language at Astra’s launch. One week later, a user was invoking the term to make the opposite point.
Other complaints have a similar shape. Developer Pankaj Kumar listed the symptoms as faster responses, worse quality, and a suspicion that OpenAI had “reduced the juice value.” Founder Saba asked OpenAI directly why she now has to “dumb it down” to get tasks completed.
“Juice value” is not an official term, and nobody has formally defined it. In this discussion, though, users are using it as shorthand for how much compute the model spends “thinking” before answering. Some suspect OpenAI quietly turned that dial down after the launch demos had already done their work.
Same-prompt comparisons added to the criticism
Some users said they tested that idea rather than relying on impressions alone. Salio and researcher Md Ismail Sojal both ran identical prompts against launch-day Astra and the current version. Both said the newer output was worse.
One post cited by Decrypt said the same prompt and the same settings produced a gap that was larger than expected when comparing Astra at launch with Astra today.
Some teams say they have already switched back
Others are not waiting for an explanation. Dax Raad, who builds the coding tool Opencode, said his team had gone back to Astra’s predecessor, GPT-5.6 Sol, because the spend doubled without enough upside to justify it.
ChatGPT user Mustafa Sahinli put it more bluntly. He said Astra now reminds him of Claude Opus 4.6 “after 1 week of release,” a jab at Anthropic’s own post-launch backlash.
Counterargument: the model did not change, users did
Not everyone thinks OpenAI deliberately weakened its new flagship. The pseudonymous user Antikythera published one of the most detailed rebuttals in the discussion and argued that the timeline is backwards.
Antikythera wrote, 「It is as dumb as it was on launch. The model is good, but the model has a lot of problems. It's lazy. Writes like a bullet-point-addict... people were overhyped on launch week, now they had time to test it and see its mistakes.」
That argument is simple: Astra may not have changed at all. Users were dazzled at launch and only later noticed that the model had always come with obvious flaws.
T3Chat founder Theo offered a similar explanation. In his view, Astra is more inconsistent than Claude Fable, which means it can generate exceptional code at one moment and obviously bad code at another. What may be happening now is that users are posting the bad examples more often because the honeymoon period is over.
Theo wrote, 「GPT-6 Astra has done incredible things I never thought a model could do. It has also done some of the stupidest things I've ever seen a model do. Generally speaking, Fable 5.1 just does what I ask.」
The same cycle hit GPT-5.6 Sol in July
This is not the first time OpenAI has faced that kind of backlash. In July, users made nearly identical complaints about GPT-5.6 Sol, the company’s previous flagship, saying its highest reasoning mode had suddenly become shallow. OpenAI executive Tibo Sottiaux denied that the model had been intentionally weakened, while confirming that the company had been experimenting with “reasoning effort,” a setting that controls how many steps a model thinks through before answering.
One reply summed up the running joke this way: closed labs release a model, it catches “some kind of disease” a few days later, and suddenly becomes dumber. Another user offered a more cynical one-line theory: 「They probably get quantized so they're not burning the companies as much money.」
Quantization means reducing the precision of a model’s internal math to lower costs, often with some trade-off in accuracy. OpenAI has never confirmed that it deliberately applies that kind of change to a model after release.
No OpenAI statement yet, while Astra still carries a higher price
So far, OpenAI has not issued a Sol-style statement about Astra.
The model still remains OpenAI’s first to cross what the company calls the critical threshold for cybersecurity risk. According to the article, that means Astra can find and chain together unknown software vulnerabilities without human guidance. That capability is restricted to vetted defenders under OpenAI’s Daybreak program.
Dumber or not, Astra’s pricing has not moved. It still costs $10 per million input tokens and $50 per million output tokens, which Decrypt said is 2.5 times what Sol charged at launch.