Codex definitely has a higher pass mark than Opus, often finds issues that Opus misses That being said, Opus is a better collaborator, it’s more explicit in what it does and is just more “fun” to work with

I also feel the same. But also it is kind of more stringent, more calculated. Like it wants to do a more accurate job.

Interesting - I even use Sonnet over codex, seems to be much better at sticking to the plan and grounding the work in other code I already have. Maybe my prompting works better with Claude than it does with codex?

do u use something like openrouter to pick & choose models? or have multiple subscriptions? i've found Claude models to be a bit lazy and prone to take shortcuts, need to steer them a lot.