Ask HN: Is GPT-6 Astra worth the 2.5x cost increase over GPT-5.6 Sol?

Comments on social media (when discounting the ironic posts) are mixed whether Astra is worth the cost increase, similar to the Fable-Opus transition.

For the novel projects I'm working on, Astra did unblock me where I was stuck with GPT 5.6 Sol, which in itself gives it value for myself, but outside of that it's not a significant improvement if at all. I have noticed Astra is more efficient so for my use case it may not be 2.5x cost.

ipaddr 3h

For the average person on here no. In fact 5.6 is probably unneeded. What DeepSeek 731 flash offers most wouldn't need anything else.

I use to want to use the best model because I wanted the highest quality output but they all produce that now. If you are trying to understand an ancient language humans don't or solve a math mystery or fixing old undocumented firmware maybe. If you are trying to create a react app or backend I wouldn't bother.

The rule of thumb is if Stack overflow could answer your question don't bother with Astra or Sol.

runjake 7h

It depends on what you're having the agent work on, but in most cases no.

speedgoose 10h

It’s pretty good at CAD.