What languages do you work with? The two models perform differently on different tech stacks. Generally as of today, GPT-6 is downright the best overall model at raw code generation / cross discipline understanding / bugfixing. The issues you might run into are straight refusals due to the security limitations, which can be quite annoying. Anthropics models are better at creative plans and generally filling in the gaps (this is your "not as smart in figuring things out"). OpenAI's models feel slightly more "crude" too for lack of a better word. You really have to be specific with your prompts and they're far less communicative also (which is a plus because they don't pepper your code with billions of pointless comments)

Money wise ChatGPT wins outright because they give you usage resets almost on a daily basis (see https://codex-resets.com/) This probably won't last, but the result of this is that I'm able to run Sol/Astra near 24/7 and by the time my weekly budget runs out I just get a reset so I'm able to continue. This IS luck based though and kind of have to time it/schedule your usage around them.

I can't really tell you whether you'll like Astra or not. It's steered the same way Sol was, so it's equally as crude but it is far better at complex understanding and optimization. So if you need that you'll be better off than with Fable.