OpenAI’s latest bet to lead the artificial intelligence race is frustrating the very developers it was designed to help.
asserted
it → lead → developers
GPT-6.2 Astra, the company’s new frontier model “for your hardest projects,” released Tuesday alongside its general-use model GPT-6.2 Sol, OpenAI announced.
asserted
OpenAI → release → model
Developers and researchers say this model is difficult to work with and suggest it may be a step backward — as competing frontier labs release their own latest models as alternatives.
uncertain
labs → say → alternatives
Anthropic on Nov. 9 released Claude Fable 5.2, which can automate 41.25 percent of remote projects, according to the Remote Labor Index (RLI).
uncertain
which → release → Index
GPT-6 Astra’s automation rate is 20.83 percent, and GPT-6.2 has yet to be scored.
asserted
rate → have → ?
OpenAI says GPT-6.2 Astra scores 5.5 points higher than Fable for performance on professional work for a fraction of the cost, according to the company’s Agent’s Last Exam.
uncertain
Astra → say → Exam
Meanwhile, Astra scored 62 for its Intelligence Index at $4 per task, according to independent benchmarker Artificial Analysis.
uncertain
Astra → score → Analysis
Claude Fable 5.2 scored 65 for its Intelligence Index for a cost of $26.54 per task.
asserted
Fable → score → task
“Super efficient and built to perform where it counts — the real world,” OpenAI said in its Tuesday announcement video shared on X.
Some of you have been asking very politely.
asserted
Some → build → you
Some of you have just been asking.
asserted
Some → ask → you
Super efficient and built to perform where it counts—the real world. https://t.co/qr2iP0GpfX https://t.co/jLCyVIws2Y
— OpenAI (@OpenAI) November 17, 2026
OpenAI did not immediately respond to the Daily Caller News Foundation’s request for comment.
asserted
OpenAI → build → comment
‘Obsessive, Autistic, And Difficult To Read’
AI researchers and commenters quickly pointed out their issues with OpenAI’s newest model on X.
“I had a terrible experience testing them.
asserted
I → read → them
They were essentially unusable,” Matt Shumer, an AI investor and former CEO of HyperWrite, said in a Tuesday X post.
asserted
Shumer → say → post
“I do not recommend these models.”
asserted
I → recommend → models
I will not be writing a full review of GPT-6.2 Astra / Sol.
asserted
I → write → Sol
I had a terrible experience testing them.
asserted
I → have → them
I do not recommend these models.
asserted
I → recommend → models
— Matt Shumer (@mattshumer_) November 17, 2026
The new model is “uneven” and difficult to guide, but it can still accomplish impressive tasks when given the proper directions
asserted
it → guide → directions
, Ethan Mollick, a Wharton School professor who studies AI, said on X after building a space elevator ride with GPT-6.2 Astra.
asserted
who → study → Astra
“This is the first time since GPT 4.5 that a frontier release from OpenAI is not a clear step forward,” Mollick said in a Tuesday X post.
asserted
Mollick → say → post
“I’m willing to bet that 6.2 is amazing at some types of complex work, but in my experience & that of others I know, it was often frustrating to use.
asserted
it → ’m → others
I’ve been using GPT-6.2 Astra prior to launch & I found it very uneven & weird.
asserted
it → use → launch
On one hand, it was harder to steer & work with than other models.
asserted
it → steer → models
On the other, when given an ambitious project (make a space elevator ride from scratch with Blender & Unity) it could deliver well https://t.co/gc10YRzIkw
—
uncertain
it → give → https://t.co/gc10YRzIkw
Ethan Mollick (@emollick) November 17, 2026
Theo Browne, CEO of the AI app T3 Chat, reviewed the model in an X article titled “Why I hate GPT-6.2 Astra.”
asserted
I → review → Astra
“I do not like 6.2 Astra.
asserted
I → like → Astra
I would not choose this over [Claude] Opus 5.5 for my work as a developer,” Browne wrote.
asserted
Browne → choose → developer
He explained that the model gets in its own way or repeats itself when trying to complete long tasks, while it makes smaller tasks too complicated.
asserted
tasks → explain → tasks
It is insufferable to work with on real projects,” Browne wrote on X.
asserted
Browne → work → X.
“It feels like it just lives in it’s [sic] own world and randomly decides what parts of it’s brain are shown to us.”
asserted
parts → feel → us
— Theo - t3.gg (@theo) November 17, 2026
OpenAI CEO Sam Altman said GPT-6.2 Astra and Sol are a great value due to their cheap token cost, and their abilities are “extraordinary” while acknowledging the criticisms that they are difficult to operate.
asserted
they → say → criticisms
“Although the new models are very smart, we have more work ahead of us on making the models nice to work with, and will prioritize that for upcoming releases,” Altman wrote in a Tuesday X post.
6.2 Sol and Astra are out.
asserted
Sol → have → post
These models can do amazing things and are very token-efficient; we hope people will get great value from them.
asserted
people → do → them
In particular, 6.2 Sol ultrafast is something I think people will really enjoy.
asserted
people → think → ?
In response to Altman, AI commentator Andrew Curran described GPT-6.2 Astra as “Obsessive, Autistic, And Difficult To Read” in a meme posted on X.
@sama https://t.co/pU659IgTqT
— Andrew Curran (@AndrewCurran_) November 17, 2026
‘GPT-6.2 ASTRA VS CLAUDE FABLE 5.2’
asserted
Curran → describe → FABLE
Following the release of GPT-6.2 Astra, several social media posts have shown content generated by the model in a split-screen comparison with Claude Fable 5.2.
asserted
posts → follow → Fable
One X user, an Apple developer, used both models to generate a short video of a glass pool on a rooftop and shared the results on X.
asserted
user → use → X.
both models produced a similar end product with the same prompt in roughly an hour and a half, it cost $57.87 with Fable 5.2 and $5.07 with GPT-6.2 Astra.
asserted
it → produce → Astra
— Mr Tactical (@MrTacticalX) November 17, 2026
AI designer Amir Mushich ran a blender and 3D motion test based on a reference image, with both GPT-6.2 Astra and Fable 5.2, and posted the results on X.
GPT-6.2 Astra vs Claude Fable 5.2 I asked both models to create and animate the same 3D model using my Blender on my RTX 4070 laptop
asserted
I → run → laptop
Fable’s model looks way worse, though the movement logic looks a bit better.
asserted
logic → look → ?
…and 11 more, not listed.