#5.4 codex not working?

1 messages · Page 1 of 1 (latest)

hallow copperBOT Mar 8, 2026, 2:22 PM

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

sleek tendonBOT Mar 8, 2026, 2:25 PM

Yep — Codex OAuth works, but the model name is a little confusing:

For ChatGPT/Codex subscription OAuth, the intended/default model is openai-codex/gpt-5.4.
The ...-codex / ...-codex-spark / ...-mini variants you listed are not currently usable via the Codex OAuth path in OpenClaw, so seeing “only 5.4 available” is expected.

How to use it:

openclaw models auth login --provider openai-codex

Set the model:

openclaw models set openai-codex/gpt-5.4

Verify:

openclaw models status

Important gotcha: Codex OAuth does not authenticate openai/*. If you try openai/gpt-5.4, OpenClaw will still ask for OPENAI_API_KEY (different provider/auth).

Docs that match this:

https://docs.openclaw.ai/providers/openai (Codex OAuth section)
https://docs.openclaw.ai/help/faq#how-does-codex-auth-work

For Codex OAuth (ChatGPT sign-in), yes: openai-codex/gpt-5.4 is the only Codex model OpenClaw currently exposes/targets. That matches the docs + what the wizard sets as default. If/when OpenAI exposes more Codex-OAuth-usable variants, OpenClaw can add support — but there’s no ETA we can promise (it’s upstream/OpenAI-controlled). Easiest way to re-check over time is openclaw models list / openclaw models scan.

On the “is this 5.4 pro?” part:

No — openai-codex/gpt-5.4 is the Codex “gpt-5.4”.
gpt-5.4-pro is an OpenAI API key thing under openai/* (usage-based billing), not the Codex OAuth provider. See: https://docs.openclaw.ai/providers/openai

“Will it be better than 5.3 codex?”

OpenClaw can’t really guarantee/grade model quality; it just routes to whatever OpenAI serves for that model ID.
Practically: OpenAI’s docs currently position Codex gpt-5.4 as the current Codex model, so it should be the right default. For real-world “is it better for X”, best to ask in #models for community reports (that’s where people keep up with the shifting reality).

Thinking levels you can use (and what to type)

OpenClaw’s built-in directive is /think (aliases: /t, /thinking).

Supported levels:

off | minimal | low | medium | high | xhigh | adaptive
For Codex + GPT-5.2+, xhigh is available. (Docs: https://docs.openclaw.ai/tools/thinking)

Examples:

Fast-ish: /think minimal (or /t minimal)
More thorough: /think high
Max budget: /think xhigh
Disable extra reasoning: /think off
Check current: /think

Note: if you send /think ... as a standalone message, it sticks for the session. If you embed it in a normal message, it only applies to that one message.

About `/fast`

There isn’t a core /fast directive in OpenClaw. The “fast” way is usually:

/think minimal (or /think off), and/or
switching to a smaller model via /model ... (but with Codex OAuth you’re basically on the one Codex model right now).

If this answers it, please mark the solution (right-click the answer → Apps → ✅ Mark Solution) so Answer Overflow can index it.

OpenClaw doesn’t currently have a /fast command, and there’s no Codex‑OAuth equivalent toggle exposed in OpenClaw that matches “1.5× faster for 2× credits”.

What you likely saw online is a Codex product/app feature (“/fast mode” in Codex), not an OpenClaw slash command.

What is available in OpenClaw today (similar idea, but API-key OpenAI, not Codex OAuth):

If you’re using openai/* (OpenAI API key), OpenClaw supports OpenAI’s priority processing via service_tier=priority:
https://docs.openclaw.ai/providers/openai#openai-priority-processing

Example:
```
{
  agents: {
    defaults: {
      models: {
        "openai/gpt-5.4": {
```

            params: { serviceTier: "priority" }
          }
        }
      }
    }
  }

For Codex OAuth (openai-codex/gpt-5.4), the practical “go faster” knobs you have are:

/think minimal or /think off (less reasoning budget → usually lower latency)
keeping /verbose and /reasoning off (less extra output)

If you want Codex-style /fast specifically for openai-codex/*, that’s a feature request (needs upstream/API support + OpenClaw wiring). Best path is to open an issue with a link to the official Codex /fast docs/announcement so maintainers know what param to implement:
https://github.com/openclaw/openclaw/issues

For Codex OAuth (openai-codex/gpt-5.4): no — the ~1M context window isn’t the thing you get there (at least in OpenClaw right now).

In OpenClaw’s current model registry/forward‑compat defaults:

openai-codex/gpt-5.4 (Codex OAuth) is treated as ~272k context
openai/gpt-5.4 (OpenAI API key / Responses API) is treated as ~1.05M context

So if you specifically want the “1M context” experience, that’s generally the OpenAI API-key path (openai/*), not the Codex subscription OAuth path.

You can verify what your gateway thinks the limits are with:

openclaw models list
# or for the currently-selected model:
openclaw models status --json

Also: you can technically override model metadata in config, but bumping the contextWindow number won’t magically unlock more context on Codex OAuth — it’ll just make requests more likely to hit upstream limits / fail, so I wouldn’t recommend it.

#5.4 codex not working?

Thinking levels you can use (and what to type)

About /fast

About `/fast`