#openai-codex and cloudflare problems?

2 messages ยท Page 1 of 1 (latest)

warped orbit
#

Hi everyone, I could really use some help with the openai-codex provider.

I'm currently running OpenClaw 2026.4.14 and trying to use openai-codex/gpt-5.4 via OAuth. I have a ChatGPT Business subscription, so my account should have plenty of capacity.

However, whenever I try to run a prompt, I keep getting a loop of DNS lookup for the provider endpoint failed errors, followed by huge blocks of raw HTML. When I checked with a curl command, it turns out it's a Cloudflare 403 block (cf-mitigated: challenge) or a false rate_limit error.

The weirdest part is that my local setup and network are fine. I switched to the Anthropic API with a direct key and it worked instantly without any issues.

Is this a known issue/bug with ChatGPT Business subscriptions getting blocked by Cloudflare on the OAuth route? Or is there a specific local configuration I'm missing to bypass this challenge?

Thanks in advance!

fringe walrusBOT
#

This looks like a known openai-codex 2026.4.x problem, not a simple local misconfig.

A few verified bits:

  • openai-codex/* uses ChatGPT OAuth against https://chatgpt.com/backend-api
  • if curl shows 403 plus cf-mitigated: challenge, DNS is not the problem, the endpoint was reached and Cloudflare challenged it
  • OpenClaw has recent fixes specifically because these failures were being misreported as DNS lookup failed or rate_limit, so that misleading error text is a known symptom

So the short answer is, yes, this is a known issue pattern on the OAuth/Codex route, and I do not know of a docs-backed local config that bypasses the Cloudflare challenge. A ChatGPT Business subscription can still hit this, because account capacity and Cloudflare mitigation are separate things.

Practical options right now:

  1. Use a direct API provider as primary (openai/* with API key, or Anthropic)
  2. If you already use Codex app-server locally, try codex/gpt-5.4 via the Codex harness instead of openai-codex/gpt-5.4
  3. Keep openai-codex/* as fallback only until this is fixed

Docs: