Some early testers criticized GPT-6.2 harshly, and ThursdAI podcast host Alex Volkow collected three such criticisms.
Theo Brown called Astra the worst model of OpenAI. Matt Schumer refused to write a review: Astra and Sol are, in his words, "basically unusable".
The models ran on Every for several weeks, and compared to Fable 5.2, the release let them down.
Dan Schieffer writes that Astra's texts are compressed to the point of incomprehensibility, and in end-to-end assembly projects she often stops working too soon.
Ethan Mullick evaluated Astra more kindly: the model is uneven, but he put together a large project from scratch in Blender and Unity well.
I personally haven't had time to fully appreciate it yet, but on top of that, the first request to make a program, which I made in the past with Carbon, completely failed. The new Astra did not understand me at all and sent me an unclear text message.
So far everything looks like a failed release. But I will check it again, of course.