Summary
- Complaints about Astra's perceived decline in performance are rampant just a week after its launch.
- Users previously noted similar issues with writing quality at the time of launch.
- The last model, Sol, experienced the same criticism in July.
Just a week after its debut, GPT-6 Astra, which had impressed users with its ability to recreate Manhattan in a game engine, is now facing backlash. Many users are expressing their disappointment on social media, questioning the model's current performance.
"Astra feels significantly dumber for me today," commented popular developer synthwavedd on X. "It was only a matter of time before The Post-Launch Lobotomy. Shame."
Myriad: Predict Nvidia stock's future. Click here to make your prediction.This sentiment is not new, as users frequently report that their favorite AI tools seem to perform worse over time. While there are often valid reasons for perceived drops in effectiveness, the volume of complaints at this stage warrants further investigation. Social media is currently filled with users expressing dissatisfaction with GPT-6 Astra.
Developer Pranjal Paliwal, who had previously praised the model, revisited the code Astra generated for him and was unimpressed with the findings.
"We don't have AGI," he stated. "We have a regression."
fml, finally had a look at the code astra wrote.
I take my words back.
We don't have AGI.
We have a regression.
How can it be so smart and so dumb at the same time!— Pranjal Paliwal (@betterclever) September 11, 2026
AGI, or artificial general intelligence, refers to a type of AI that can perform any intellectual task that a human can. OpenAI's president had mentioned this concept during Astra's launch, but a week later, users are using it to describe the opposite.
The complaints are consistent across the board. Developer Pankaj Kumar enumerated the issues: quicker responses, poorer quality, and a belief that OpenAI has "reduced the juice value." Founder Saba directly questioned OpenAI about the need to "dumb it down" to achieve results.
"Juice value" is not a formally defined term but is understood to refer to the computational effort the model expends before providing an answer, with users suspecting OpenAI may have toned this down after the initial launch excitement subsided.
Some users conducted tests. Salio and researcher Md Ismail Sojal both executed the same prompt on Astra from launch day and the current version, yielding inferior results from the latter.
Today’s GPT-6 Astra output looks worse.
GPT-6 Astra at launch vs GPT-6 Astra today.- Tibo said they have compute
- Same prompt, Same settings.
- even I ran the exact same prompt on GPT-6 Astra at launch and again today.
- The difference is bigger than I expected.…— Md Ismail Šojal 🕷️ (@0x0SojalSec) September 11, 2026
Some users are reverting to previous versions. Dax Raad, who develops the coding tool Opencode, revealed that his team has returned to using Astra's predecessor, GPT-5.6 Sol, finding the benefits of Astra not worth the increased costs. ChatGPT user Mustafa Sahinli bluntly stated that Astra makes him feel like Claude Opus 4.6 "after 1 week of release," referencing a similar post-launch backlash faced by Anthropic's model.
However, not all users believe OpenAI intentionally downgraded the new model. A user named Antikythera provided a detailed counterargument, suggesting that the model was no worse than during its launch. "The model is good, but it has many 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," he wrote.
In essence, he argues that nothing has changed; users simply became more critical after the initial excitement wore off.
T3Chat founder Theo echoed a similar sentiment, noting that Astra displays greater inconsistency than Claude Fable, producing both excellent and poor results at times. He suggested that the increase in negative feedback is due to users sharing subpar outputs more frequently now that the initial novelty has faded.
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. pic.twitter.com/X2UwI7tDnD
— Theo - t3.gg (@theo) September 8, 2026