Two things I noticed about Codex vs Claude:
- Codex usage is metered separately - it does not come out of your normal ChatGPT bucket.
- Hitting the Codex limit does not kill active jobs. In-flight work finishes; new work gets blocked.
I had two sessions running side by side. One finished, and when I tried to run another prompt, I hit a usage limit. The other kept going for another 5 minutes until it finished.