Codex API latency is ~2x slower in OAuth harness vs Codex CLI for same task/model
Resolved 💬 2 comments Opened Mar 4, 2026 by swarm-agent Closed Mar 4, 2026
💡 Likely answer: A maintainer (github-actions[bot], contributor)
responded on this thread — see the highlighted reply below.
What issue are you seeing?
It seems Codex is much slower by a factor of at least 2x via the API using Oauth in other platforms (even in our personal testing we have websockets enabled and it's still fairly slow).
What steps can reproduce the bug?
Use Codex (I'm using 5.3 xhigh) side by side to do the same task on Oauth on a different harness that isn't Codex with Codex doing the same task, it's much faster.
What is the expected behavior?
To be as fast as Codex
Additional information
We're trying to push Codex to the limit in our harness, can you guys let us know if there is anything we can do, it may be we overlooked something but I'm not quite sure as getting it working via API was quite complicated!
We would really like to push Codex to its limits in our harness with its added speed and capabilities.
2 Comments
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
I'm confused by your bug report. Codex is a harness. When you say "Codex API", are you referring to the responses API used to access OpenAI-hosted models like
gpt-5.2orgpt-5.3-codex?If you're writing your own harness and want to understand how the Codex harness works, it is open source. Just clone the repo and look at the code (or use Codex to ask it questions about the code base).