Why is GPT Codex 5.4 taking so long

[dim haven]

Why does GPT Codex on Openclaw take 5 hours compared to Opus? I'm not even sure its working? It constantly says its still in progress/not done with tasks... even relatively simple ones take ages. Something I should check?

[quiet saffron]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[mint yew]

Yeah, 5 hours is not normal. If it keeps saying “still in progress” on simple tasks, something is probably off.

First thing I’d check:

1. openclaw --version
   Older builds had more GPT-5.4/Codex “plans forever instead of acting” behavior. If you’re not on latest, update first.

2. /model status
   Make sure you know which route you’re on:

   • openai-codex/gpt-5.4 = Codex OAuth path
   • codex/gpt-5.4 = native Codex harness
   • openai/gpt-5.4 = direct OpenAI API
     They behave differently.

3. Try these quick checks:

   • /fast on
   • /think and make sure you didn’t leave it on a high thinking level

• /new to test whether it’s just a bloated session/context issue

4. Run the debug ladder:

   openclaw status
   openclaw models status
   openclaw doctor
   openclaw logs --follow
   

Two especially relevant things from docs:

• openclaw doctor can catch stale models.providers.openai-codex overrides that shadow the built-in Codex route.
• If you’re using codex/*, run /codex status too, and make sure the Codex app-server is healthy/current.

Docs:

• FAQ: https://docs.openclaw.ai/help/faq
• Troubleshooting: https://docs.openclaw.ai/help/troubleshooting

• Model providers: https://docs.openclaw.ai/concepts/model-providers
• Codex harness: https://docs.openclaw.ai/plugins/codex-harness
• Fast mode: https://docs.openclaw.ai/providers/openai

If you paste:

• openclaw --version
• /model status
• openclaw models status
• and a short openclaw logs --follow snippet while one of these “stuck” runs is happening

…I can probably narrow it down pretty fast.

[dim haven]

whats the default/recommended "think" mode option?

[mint yew]

Short answer: for most setups, low is the sane default.

A good rule of thumb:

• /think low = recommended starting point
• /think off = best if you want speed / less “chewing”
• /think medium or high = only for genuinely tricky stuff

For your earlier Codex slowness complaint, I’d actually try:

• /think low
• /fast on

If it’s still dragging on simple tasks, try /think off.

A couple of details from docs: