Claude Code still gets a bit sluggish when executing long and complex projects. One head is good, but two are better, I thought, and after the experiments above, I decided to finally marry them.
Currently, I have a $200 max-package of Claude Code paid for and a $20 package of ChatGPT, so I now strictly do the following:
I tell my Claude - open my Codex in a neighboring tab at max settings and use it as an independent consultant, feeding it each block of tasks for evaluation and then for verification.
Of course, I had Codex open in the terminal beforehand with all accesses and with the model gpt-5.6-sol (reasoning xhigh, summaries auto).
Thanks to this, I have already caught a bunch of shortcomings made by Claude in my two long projects! Highly recommend!
Since Codex now regularly resets limits (oh miracle), my $20 package is enough to get a huge amount of work done! Use it while it lasts.
