Today OpenAI dropped an official Codex plugin for Claude Code, and it’s honestly a surprisingly strong combo. It’s not like this is brand-new tech—people have already been mixing Codex + Claude Code—but the plugin makes it dead simple to pull Codex into the same workflow. The big “why” is code reviews: I’ve been seeing more and more Claude Code users bring Codex in as a second pair of eyes, and the head of DevEx at OpenAI basically confirmed that’s exactly how the community is using it. In the video I compare Opus 4.6 vs GPT-5.4 benchmarks (Opus barely edges SWE-bench, but GPT-5.4 leads in a bunch of other coding benchmarks), then I go beyond the charts and test the feel. I scraped opinions from X/Reddit and it lines up with what I’ve noticed: Claude Code can over-engineer, get token hungry, and drift on long runs—then miss its own edge cases. Codex is great at auditing and catching issues, but people complain it’s more rigid and not as strong at planning or asking the right questions. So I set up the plugin, run an adversarial review on a real project, then do a head-to-head “build me a game” prompt. Codex shipped a more polished one-shot UI, and then I fed Codex’s review back into Opus to implement fixes. The takeaway: don’t marry one tool—use Claude Code to plan/build, and use Codex to pressure-test and review.