Shumer, Shipper and Theo pan GPT-6.2 on launch day
Three of the testers founders copy their model defaults from all came out against OpenAI's new release the moment the embargo lifted.
OpenAI announced that GPT-6.2 Astra and Sol are now available, describing them as "Super efficient and built to perform where it counts". Within hours, three people who had been testing the models before launch posted reviews, and all three were negative.
Matt Shumer said he would not be writing a full review because he had a terrible experience testing them. His words were that the models were "essentially unusable" and that he does not recommend them. He later said he would make a video instead, after enough people asked.
Dan Shipper, who says he and his team at @every tested the models for the last few weeks, broke his verdict into three parts. On writing, he found Astra 6.2 a step back from Astra 6, trained to be so concise and actionable that its writing is hard to understand, giving short answers where long ones are appropriate. On coding, he found performance similar to Astra 6 on most of their benchmarks, but in real tasks the model seemed to stop too early, which he said made it frustrating on end to end builds. On design, he said it has good default instincts but struggles to vary them, producing output with a noticeable house style. He called it a disappointing release compared with Fable 5.2, the model Shumer used to build a voxel engine from a single prompt earlier.
Theo, posting as Theo - t3.gg, said he had been testing 6.2 Astra for a while and thinks it is the worst model OpenAI has ever released. He said he would share more soon but is currently sick in bed with a 102 degree fever, so he will be slow.
Why they all landed at once
The simultaneous timing set off the usual theories in replies, and Shumer answered them directly. The reviews landed together because there was an embargo. They were all negative, he said, because the model is bad. He added that there is no conspiracy and the testers are just giving their opinions.
The business read is that this is three testers' impressions, not benchmark results, and none of them published numbers alongside the verdicts. Shipper's team at least ran their own benchmarks and reported coding performance roughly flat versus Astra 6, with the complaint sitting in real task behaviour rather than scores. But these are the accounts a lot of founders watch before switching the model behind their product, which makes a coordinated wall of thumbs down on day one a harder launch than a quiet one.
Why are the reviews all negative: the model is bad.
I had a terrible experience testing them. They were essentially unusable.
I think this is the worst model OpenAI has ever released.
Steal the numbers, not the vibes
One email a day. Under three minutes. Real MRR, real pricing, real launches from founders building with AI.

