The biggest difference I have noticed between Claude and Codex/GPT5.6

The biggest difference I have noticed between Claude Opus/Fable and Codex GPT5.6/any model is Codex seems pretty content to just waste time looking like it is doing things without actually doing things. It does not seem to be outcome-oriented. I threw a big coding project at it this weekend using a mix of sol/terra/luna delegated tasks at it and it will spend HOURS just doing prep work. Then maybe write some code, then spend 60-70% of it's time testing and verifying. It easily spends 20% of it's time coding