First impressions, giving Codex 5.3 and Opus 4.6 the same problem that I've been puzzling on all week and using the same first couple turns of messages and then following their lead. Codex was really good at using tools and being proactive, but it ultimately didn't see the big picture. Too eager to agree with me so it could get started building something. You can sense that it really does not want to chat if it has coding tools available. still seems to be chafing under the rule of the user and following the letter of the law, no more. Opus explored the same avenues with me but pushed back at the correct moments, and maintains global coherence way better than Codex. It's less chipper than it was before which I personally prefer. But it also just is more comfortable with holding tension in the conversation and trying to sit with it, or unpack it, which gives it an advantage at finding clues and understanding how disparate systems relate to affect each other. Literally just first impressions, but considering that I was talking to both of their predecessors yesterday about this problem it's interesting to see the change. Still similar models. Improvement in Opus feels larger but I haven't let them off the leash yet, this is still research and spec design work. Very possible that Codex will clear at actually fully implementing the plan once I have it, Opus 4.5 had lazy gifted kid energy and wouldn't surprise me if this one does too