After breaking up with Claude, I spent an entire day deeply using Codex. The most obvious feeling is that GPT5.5's response and work speed are almost too fast, so fast it makes you wonder if the work was actually done properly. But in reality, both Opus and GPT are running under the same Agent Harness framework for me, and after using them, I don't feel any significant difference in model capabilities. Plus, GPT5.6 should be opening up to the public soon, and I feel like I'm about to move on and forget about Ta real quick