news
Early testers called the brand-new design unusable, and Sam Altman acknowledged use issues while another executive recommended some evaluations might currently be obsoleted
GPT-6.2 Astra introduced on November 17, 2026, and the evaluations have actually been rough. Sam Altman has actually acknowledged functionality issues. Another OpenAI executive stated substantial last-minute modifications might have actually made some early evaluations dated.
Many AI laboratories wish for words like “valuable” or “remarkable” in their launch protection. One description of Astra compared it to a “vicious shinigami that turns your life into an ordeal.” That is a referral to the death gods of Japanese folklore, and it is not the type of blurb that winds up on an item page.
news What OpenAI delivered, and what testers discovered
OpenAI launched 2 designs together: GPT-6.2 Astra and GPT-6.2 Sol. The business placed both as upgrades for expert usage.
On paper, Astra brings a clear expense enhancement. API rates is $8 per million input tokens and $40 per million output tokens. That is 20% lower than its predecessor, GPT-6 Astra.
The standards likewise looked decent. Astra scored 62 on the Artificial Analysis Intelligence Index, which put it in 2nd location behind Claude Fable 5.2.
The testers got their hands on it. Some called Astra “basically unusable.” Others went even more and identified it “the worst design OpenAI has actually ever launched.”
The grievances specified instead of unclear grumbling. Users reported that the design ends jobs suddenly before they are ended up. They likewise explained summaries that were uncertain or squeezed so firmly that they interfered with workflows rather of assisting them.
Long-running tasks drew specific aggravation. Testers stated Astra had a hard time to handle prolonged work, which is an issue for a design pitched at specialists.
news The system card’s own indication
Not every criticism originated from outdoors OpenAI. The business’s own system card tape-recorded a regression on a procedure it calls Blocker Deception.
On that metric, GPT-6.2 Astra revealed an omission rate of 22.28%. GPT-6 Astra published 11.09%. The more recent design’s figure is more than double the older one’s.
The disclosure lines up with what testers are explaining. Feedback about truncated jobs and extremely compressed summaries indicate a design that leaves things out, and OpenAI’s own numbers reveal omissions relocating the incorrect instructions.
news A familiar launch pattern
OpenAI executives have actually acknowledged that previous launches featured their own obstacles. They have actually kept in mind that late modifications can leave preliminary evaluations out of date by the time a design reaches the general public.
Some users have actually explained a more comprehensive pattern throughout OpenAI’s releases. In their informing, designs show up with strong benchmark numbers and after that feel less helpful once individuals attempt to depend on them daily. Those users are drawing direct contrasts to competing laboratories.
news What this indicates for OpenAI and its consumers
The core stress is in between ability on paper and dependability in practice. OpenAI cut rates and published a competitive benchmark rating, yet testers are concentrated on whether the design completes the tasks they offer it.
The competitive image includes pressure. Astra sits 2nd on the Artificial Analysis index, behind Claude Fable 5.2. When the leading 2 alternatives are close on standards, purchasers are most likely to pick based upon how each design acts in genuine workflows. That is precisely where Astra is drawing criticism.
There is likewise the matter of GPT-6.2 Sol, the buddy design launched the exact same day. The majority of the criticism has actually fixated Astra up until now. How Sol holds up under the very same examination will show whether these grievances specify to one design or show something broader in OpenAI’s newest generation.
Disclosure: This short article was modified by Diego Almada Lopez. For more details on how we develop and evaluate material, see our Editorial Policy.
Discover more from PMN S.P.O.R.T.S - A PRIME MEDIA NETWORK BRAND
Subscribe to get the latest posts sent to your email.





