{"detail":"The 'gpt-5.3-codex' model is not supported when using Codex with a ChatGPT account."} | Friends of the Crustacean 🦞🤝 | Page 1

compact flumeBOT Mar 10, 2026, 8:39 AM

#

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

faint sedgeBOT Mar 10, 2026, 8:41 AM

#

That error means you’re authenticated via Codex OAuth / ChatGPT account, but you’ve selected a model (gpt-5.3-codex) that isn’t available under that auth type. A ChatGPT/Codex login ≠ OpenAI API access for arbitrary models.

What to do:

Confirm what OpenClaw thinks you can use

Run (on the gateway host):
openclaw --version
openclaw models status
Or in-chat: /model status and /model list

Fix the model you’re using for the daily summary

If your daily summary is a cron job, it likely has model: "gpt-5.3-codex" set somewhere. Change it to a model that shows up in /model list.
If you’re just running it in a normal chat session, switch with /model and pick one from the list.

If you specifically want to use gpt-5.3-codex:

You’ll need to set up OpenAI API key auth (provider openai), not Codex OAuth.

#

Docs refs:

If you paste the output of openclaw models status (it redacts secrets), I can tell you exactly which config entry/job is pointing at gpt-5.3-codex.

faint sedgeBOT Mar 10, 2026, 9:06 AM

#

Yep — your models status explains it perfectly:

Default: openai-codex/gpt-5.3-codex
Auth for openai-codex: OAuth (ChatGPT/Codex login)

…and OpenAI is telling you that gpt-5.3-codex isn’t usable with that OAuth login type, so every run that uses the default model will fail (including your daily summary).

Fix (pick one)

Option A: Stay on Codex OAuth, switch to a Codex model that OAuth supports

The FAQ notes the wizard default should be openai-codex/gpt-5.4, so try:

openclaw models set openai-codex/gpt-5.4

Then re-run your daily summary.

#

(You can confirm with openclaw models status afterward.)
Docs: https://docs.openclaw.ai/cli/models and https://docs.openclaw.ai/help/faq#how-does-codex-auth-work

Option B: Use Anthropic for the daily summary (you already have an Anthropic API key)

If the summary is just normal “assistant writing”, you can set your default model to an Anthropic one (pick from openclaw models list) and only switch to Codex when you’re doing coding.

If you set the model inside a cron job

If your “daily summary” is a cron and it has its own model: override, changing the default won’t help — you’ll need to update the cron’s model too. Tell me whether your summary is a cron or a manual prompt, and I’ll point you at the exact place to change it.

If Option A fixes it, please mark the solution in the thread (Apps → ✅ Mark Solution) so Answer Overflow indexes it.

#

You don’t have to remove Anthropic — it won’t be used unless you pick an anthropic/... model (or add it as a fallback). But if you want it gone for cleanliness/safety, you can remove the stored Anthropic credential.

From your models status, the Anthropic key is coming from:

~/.openclaw/agents/main/agent/auth-profiles.json (profile anthropic:default)

Safe removal (manual, with backup)

Make a backup:

cp ~/.openclaw/agents/main/agent/auth-profiles.json ~/.openclaw/agents/main/agent/auth-profiles.json.bak

Edit the file and delete the Anthropic profile entry (look for anthropic:default / provider: "anthropic" / api_key under anthropic).
Make sure the JSON stays valid (commas/brackets).
Restart + verify:

#

openclaw gateway restart
openclaw models status

Also check (optional)

If you ever put ANTHROPIC_API_KEY in ~/.openclaw/.env, remove it there too (then restart). In your output it looks like Anthropic is not coming from shell env, so it’s almost certainly just the auth-profiles file.

If you paste the redacted contents of auth-profiles.json (remove the actual key string), I can tell you exactly which block to delete so you don’t have to guess.

hot quest Mar 10, 2026, 10:34 AM

#

{"detail":"The 'gpt-5.4' model is not supported when using Codex with a ChatGPT account."}

faint sedgeBOT Mar 10, 2026, 10:35 AM

#

That new error is the same underlying issue: your current auth is “openai-codex via ChatGPT/OAuth”, and OpenAI is rejecting the model you’re trying to run under that login.

In other words: it’s not an OpenClaw bug — it’s model entitlement vs auth type. Some OpenAI models simply can’t be used with “ChatGPT account” (Codex OAuth) and require an API key (or your account needs actual Codex access).

Fastest way to get unstuck: auto-detect a working Codex model

Run a scan for the openai-codex provider and let OpenClaw set something that actually works:

openclaw models scan --provider openai-codex --set-default

(Heads up: “scan” does real probes/requests.)

Then check:

#

openclaw models status

If scan can’t find any working model

Then your OpenAI account likely doesn’t have Codex access via OAuth (even if you can log into ChatGPT). In that case you have two real options:

Use an API-key provider for the daily summary (OpenAI API key, or keep Anthropic since you already have it), or
Enable Codex on that OpenAI account (whatever OpenAI currently requires), then re-login.

If you paste the output of:

openclaw models scan --provider openai-codex --json

I can tell you whether it found any usable Codex model for your OAuth login, or whether you’ll need to switch to API-key auth.

hot quest Mar 10, 2026, 10:40 AM

#

openclaw@openclaw:~$ openclaw models scan --provider openai-codex --json
Error: Missing OpenRouter API key. Set OPENROUTER_API_KEY to run models scan.

faint sedgeBOT Mar 10, 2026, 10:41 AM

#

openclaw models scan isn’t a “scan my Codex account” command — it specifically scans OpenRouter’s free model catalog, so it will ask for an OpenRouter API key (OPENROUTER_API_KEY). That’s why you got that error. (My bad for suggesting scan for openai-codex.)

For your actual problem (Codex OAuth):

You’re logged in with openai-codex OAuth (“ChatGPT account”)
OpenAI is rejecting both gpt-5.3-codex and gpt-5.4 under that auth

That strongly suggests your OpenAI login does not have Codex model entitlement (or OpenAI has restricted what that OAuth can call). In that case, there may be no openai-codex/... model that will work until the account has Codex access.

What to do next (2 quick checks)

See what OpenClaw thinks is selectable:

openclaw models list

If you see any openai-codex/... entries, try setting one of those:

#

openclaw models set <whatever-you-saw>

If everything openai-codex/... fails anyway, you need to switch auth type:

Use an API key provider (OpenAI API key → openai/..., or keep using your existing Anthropic key), because ChatGPT subscription/OAuth isn’t the same as API access.

Docs:

Codex OAuth support/notes: https://docs.openclaw.ai/help/faq#do-you-support-openai-subscription-auth-codex-oauth
Models CLI: https://docs.openclaw.ai/cli/models

If you paste the output of openclaw models list (no secrets in that output), I can tell you whether there’s any openai-codex/... option worth trying—or whether you’re forced to use API-key auth for the daily summary.

hot quest Mar 10, 2026, 10:42 AM

#

openclaw@openclaw:~$ openclaw models list

🦞 OpenClaw 2026.3.8 (3caab92)
If it's repetitive, I'll automate it; if it's hard, I'll bring jokes and a rollback plan.

Model Input Ctx Local Auth Tags
openai-codex/gpt-5.4 text+image 266k no yes default,configured

faint sedgeBOT Mar 10, 2026, 10:46 AM

#

Perfect, that output is the smoking gun:

openclaw models list shows only:

openai-codex/gpt-5.4 (Auth: yes)

…but OpenAI is still returning:

"The 'gpt-5.4' model is not supported when using Codex with a ChatGPT account."

So right now your Codex OAuth token is valid, but your OpenAI account doesn’t have permission to use that Codex model. And since that’s the only openai-codex/... model OpenClaw knows about, there’s nothing else to switch to within Codex OAuth.

What you can do

Use an API key provider instead (recommended for your daily summary)

Either OpenAI API key (openai/...) or keep using your Anthropic API key.
Easiest path: run the interactive picker and choose a non-Codex provider/model:

#

  openclaw configure --section model

(or openclaw models auth add to add an API-key provider, then openclaw models set ...)

If you specifically want Codex OAuth
You need to sign in with an OpenAI account that actually has Codex access/entitlement (whatever OpenAI currently requires). A normal ChatGPT account often won’t cut it, which is exactly what that error is telling you.

Docs refs:

Codex OAuth note: https://docs.openclaw.ai/help/faq#do-you-support-openai-subscription-auth-codex-oauth
Models setup: https://docs.openclaw.ai/cli/models

If you tell me “I want to use OpenAI via API key” vs “I want to use Anthropic” (and whether you’re OK keeping Anthropic installed), I’ll give you the exact minimal commands to switch your default model so the daily summary works.

hot quest Mar 10, 2026, 11:06 AM

#

i want to use Google Gemini API key

faint sedgeBOT Mar 10, 2026, 11:08 AM

#

Cool — Gemini API key is the straightforward path.

1) Put your key where the Gateway will actually see it

Since your models status showed Shell env: off, don’t rely on export ... in your terminal. Put it in the global OpenClaw env file instead:

nano ~/.openclaw/.env

Add:

GEMINI_API_KEY=YOUR_KEY_HERE

(save + exit)

#

Docs on env loading: https://docs.openclaw.ai/help/environment

2) Restart the gateway

openclaw gateway restart

3) Pick a Google model + set it as default

List Google models:

openclaw models list --provider google --all

Then set one (copy the exact google/... id from the list):

#

openclaw models set google/<model-from-the-list>

Finally verify:

openclaw models status

Once your default is google/..., your “daily summary” will stop trying to use Codex OAuth (which is what’s causing the “not supported with ChatGPT account” errors).

If models list --provider google --all comes back empty, paste that output + openclaw models status and we’ll see why the key isn’t being picked up.

hot quest Mar 12, 2026, 5:52 PM

#

which openai Oauth is supported?

faint sedgeBOT Mar 12, 2026, 5:54 PM

#

OpenClaw only supports one OpenAI OAuth-style login today:

OpenAI Code (Codex) OAuth → provider openai-codex (ChatGPT sign-in)

It does not support a generic “OpenAI API via OAuth” flow for the normal openai/... models — for those you need an OpenAI API key (OPENAI_API_KEY) and use provider openai.

Also important (based on the errors you were seeing earlier): even if OAuth login works, your OpenAI account still needs Codex entitlement for the Codex model(s), otherwise OpenAI returns “model not supported with a ChatGPT account”.

Docs:

https://docs.openclaw.ai/help/faq#do-you-support-openai-subscription-auth-codex-oauth
https://docs.openclaw.ai/concepts/model-providers (see “OpenAI Code (Codex)” vs “OpenAI”)
https://docs.openclaw.ai/providers/openai

hot quest Mar 12, 2026, 5:54 PM

#

do i need to be subscribed user?

faint sedgeBOT Mar 12, 2026, 5:57 PM

#

Not for OpenClaw itself — but yes/no depends on which auth path you want:

OpenClaw: no subscription required. You can run it purely with API keys (pay‑per‑use) from Anthropic/OpenAI/Google/etc.
Docs: https://docs.openclaw.ai/help/faq#do-i-need-a-claude-or-openai-subscription-to-run-this
OpenAI OAuth in OpenClaw (Codex OAuth / openai-codex): you need an OpenAI account that has Codex access/entitlement. A regular ChatGPT account/plan often doesn’t include that, which matches the “model not supported when using Codex with a ChatGPT account” errors you hit.
Google Gemini (API key / google provider): you don’t need a “subscription”, but you do need a Gemini API key (and Google may require enabling the API / billing depending on your region/account).

If you tell me which one you want to rely on day-to-day (Codex OAuth vs Gemini API key vs OpenAI API key), I can suggest the cleanest config so your daily summary stops failing.

hot quest Mar 12, 2026, 6:19 PM

#

Error: channels.telegram.botToken: unresolved SecretRef "env:default:TELEGRAM_BOT_TOKEN". Resolve this command against an active gateway runtime snapshot before reading it.

faint sedgeBOT Mar 12, 2026, 6:20 PM

#

That means your config has Telegram enabled and channels.telegram.botToken is set to a SecretRef pointing at an env var:

env:default:TELEGRAM_BOT_TOKEN

…but the Gateway runtime can’t resolve it because TELEGRAM_BOT_TOKEN isn’t present in the Gateway’s environment (or you’re running the command without a running gateway snapshot).

Fix (if you want Telegram)

Put the token into the global env file (so the service sees it too):

nano ~/.openclaw/.env

Add:

#

TELEGRAM_BOT_TOKEN=123456:abc...

Restart + reload secrets:

openclaw gateway restart
openclaw secrets reload

(Then openclaw channels status / openclaw logs --follow if you want to verify.)

Docs: https://docs.openclaw.ai/channels/telegram (see “Env fallback: TELEGRAM_BOT_TOKEN=…”)

If you don’t want Telegram

Disable/remove the Telegram channel config (so it stops trying to resolve that token). The simplest is setting channels.telegram.enabled: false or removing the channels.telegram block.

#

Tell me which case you want (use Telegram vs disable it), and I’ll suggest the cleanest change path.

hot quest Mar 12, 2026, 6:23 PM

#

the telegram bot token is already on the .env

faint sedgeBOT Mar 12, 2026, 6:24 PM

#

If it’s already in a .env file, then one of these is almost always true:

it’s in the wrong .env file (not the one the Gateway reads), or
the Gateway is running under a different state dir/user/profile, so it’s reading a different ~/.openclaw/.env, or
the Gateway hasn’t been restarted / secrets reloaded since you added it, or
the variable name doesn’t match exactly (TELEGRAM_BOT_TOKEN).

Here’s the quickest way to pinpoint which:

A) Confirm you edited the right file + key name (don’t paste the value)

ls -la ~/.openclaw/.env
grep -n '^TELEGRAM_BOT_TOKEN=' ~/.openclaw/.env

#{"detail":"The 'gpt-5.3-codex' model is not supported when using Codex with a ChatGPT account."}