Bug type
Regression (worked before, now fails)
Beta release blocker
No
Summary
OpenAI/Codex device pairing succeeds and GPT-5.5 is usable through the Codex runtime.
However, after successful authentication:
- the configuration wizard never displays a model picker,
- the wizard immediately returns to the main configuration menu,
- model catalog state becomes internally inconsistent,
- the active model is not present in the allowed-model list.
The issue has persisted across multiple OpenClaw releases.
Steps to reproduce
Step 1
Run:
openclaw config
Step 2
Choose:
Model
OpenAI
ChatGPT/Codex Device Pairing
Step 3
Complete the browser authentication flow.
Step 4
Observe:
OpenAI device code complete
Step 5
Observe that OpenClaw immediately returns to the configuration menu instead of displaying a model-selection screen.
Evidence: Authentication Works
Authentication succeeds.
openclaw models status --json
shows:
"providersWithOAuth": [
"openai-codex (1)"
]
and
{
"provider": "openai-codex",
"status": "ok"
}
OAuth credentials are present and usable.
Evidence: Runtime Works
- GPT-5.5 executes successfully.
- Control UI session output reports:
Model: gpt-5.5
Runtime: OpenAI Codex
Therefore this does not appear to be an authentication failure.
Evidence: Catalog State Becomes Inconsistent
Current model list
Model Input Ctx Local Auth Tags
codex/gpt-5.5 text 195k no yes default
openai/gpt-5.5 text 195k no yes configured
openai/gpt-5.3-codex text 195k no yes configured
codex/gpt-5.2 text+image 266k no yes
codex/gpt-5.3-codex text+image 266k no yes
codex/gpt-5.4 text+image 266k no yes
codex/gpt-5.4-mini text+image 266k no yes
Current model status
"defaultModel": "codex/gpt-5.5",
"resolvedDefault": "codex/gpt-5.5"
but:
"allowed": [
"openai/gpt-5.5",
"openai/gpt-5.3-codex"
]
The active default model is not present in the allowed-model list.
Why This Appears To Be A Bug
The following simultaneously hold true:
- OAuth authentication succeeds.
- GPT-5.5 runs successfully.
- The configuration wizard does not display a model picker.
- The active model differs from the allowed-model list.
- Catalog entries appear internally inconsistent.
This suggests a problem in model discovery, catalog normalization, provider mapping, or model-selection UI flow rather than a user configuration error.
Failure mode has been persistent for 2 instances on separate infrastructure across openclaw v26.5.22 / 5.26 / 5.27 / 5.28 (only release versions tested, maybe earlier versions affected)
Expected behavior
After successful authentication, openclaw config should display the previously working model picker.
Actual behavior
After successful authentication, openclaw config falls back to the main menu without ever displaying the previously working model picker.
OpenClaw version
2026.5.28
Operating system
Debian 13
Install method
npm
Model
codex plan / GPT-5.5 and GPT-5.3-codex tested
Provider / routing chain
openclaw-codex
Additional provider/model setup details
No response
Logs, screenshots, and evidence
openclaw@dellserver:~$ openclaw models status --json
{
"configPath": "/home/openclaw/.openclaw/openclaw.json",
"agentDir": "/home/openclaw/.openclaw/agents/main/agent",
"defaultModel": "codex/gpt-5.5",
"resolvedDefault": "codex/gpt-5.5",
"fallbacks": [],
"imageModel": null,
"imageFallbacks": [],
"aliases": {},
"allowed": [
"openai/gpt-5.5",
"openai/gpt-5.3-codex"
],
"auth": {
"storePath": "/home/openclaw/.openclaw/agents/main/agent/auth-profiles.json",
"shellEnvFallback": {
"enabled": false,
"appliedKeys": []
},
"providersWithOAuth": [
"openai-codex (1)"
],
"missingProvidersInUse": [],
"runtimeAuthRoutes": [],
"providers": [
{
"provider": "codex",
"effective": {
"kind": "synthetic",
"detail": "codex-app-server"
},
"profiles": {
"count": 0,
"oauth": 0,
"token": 0,
"apiKey": 0,
"labels": []
},
"syntheticAuth": {
"value": "plugin-owned",
"source": "codex-app-server",
"credential": "codex-app-server",
"mode": "token"
}
},
{
"provider": "openai-codex",
"effective": {
"kind": "profiles",
"detail": "~/.openclaw/agents/main/agent/auth-profiles.json"
},
"profiles": {
"count": 1,
"oauth": 1,
"token": 0,
"apiKey": 0,
"labels": [
"openai-codex:g...e@gmail.com=OAuth (g...e@gmail.com)"
]
},
"syntheticAuth": {
"value": "plugin-owned",
"source": "codex-app-server",
"credential": "codex-app-server",
"mode": "token"
}
}
],
"unusableProfiles": [],
"oauth": {
"warnAfterMs": 86400000,
"profiles": [
{
"profileId": "openai-codex:g...e@gmail.com",
"provider": "openai-codex",
"type": "token",
"status": "static",
"source": "store",
"label": "openai-codex:g...e@gmail.com (g...e@gmail.com)"
}
],
"providers": [
{
"provider": "openai-codex",
"status": "ok",
"profiles": [
{
"profileId": "openai-codex:g...e@gmail.com",
"provider": "openai-codex",
"type": "token",
"status": "static",
"source": "store",
"label": "openai-codex:g...e@gmail.com (g...e@gmail.com)"
}
],
"effectiveProfiles": [
{
"profileId": "openai-codex:g...e@gmail.com",
"provider": "openai-codex",
"type": "token",
"status": "static",
"source": "store",
"label": "openai-codex:g...e@gmail.com (g...e@gmail.com)"
}
]
}
]
}
}
}
openclaw@dellserver:~$ openclaw models list
OpenClaw 2026.5.28 (e932160) — Shell yeah—I'm here to pinch the toil and leave you the glory.
Model Input Ctx Local Auth Tags
codex/gpt-5.5 text 195k no yes default
openai/gpt-5.5 text 195k no yes configured
openai/gpt-5.3-codex text 195k no yes configured
codex/gpt-5.2 text+image 266k no yes
codex/gpt-5.3-codex text+image 266k no yes
codex/gpt-5.4 text+image 266k no yes
codex/gpt-5.4-mini text+image 266k no yes
openclaw@dellserver:~$
Impact and severity
GPT-5.5 remains usable, but the model picker cannot be invoked
Additional information
This issue has been observed across multiple releases and was not resolved by upgrading to 2026.5.28.
The behavior appears related to the interaction between:
openai/*
codex/*
openai-codex
model/provider routing.
However, runtime inference itself appears healthy.
tested settings manually:
openclaw config set agents.defaults.model.primary openai/gpt-5.5
and
openclaw config set agents.defaults.model.primary codex/gpt-5.5
with the same missing-picker behavior.
Bug type
Regression (worked before, now fails)
Beta release blocker
No
Summary
OpenAI/Codex device pairing succeeds and GPT-5.5 is usable through the Codex runtime.
However, after successful authentication:
The issue has persisted across multiple OpenClaw releases.
Steps to reproduce
Step 1
Run:
openclaw config
Step 2
Choose:
Model
OpenAI
ChatGPT/Codex Device Pairing
Step 3
Complete the browser authentication flow.
Step 4
Observe:
OpenAI device code complete
Step 5
Observe that OpenClaw immediately returns to the configuration menu instead of displaying a model-selection screen.
Evidence: Authentication Works
Authentication succeeds.
openclaw models status --json
shows:
"providersWithOAuth": [
"openai-codex (1)"
]
and
{
"provider": "openai-codex",
"status": "ok"
}
OAuth credentials are present and usable.
Evidence: Runtime Works
Model: gpt-5.5
Runtime: OpenAI Codex
Therefore this does not appear to be an authentication failure.
Evidence: Catalog State Becomes Inconsistent
Current model list
Model Input Ctx Local Auth Tags
codex/gpt-5.5 text 195k no yes default
openai/gpt-5.5 text 195k no yes configured
openai/gpt-5.3-codex text 195k no yes configured
codex/gpt-5.2 text+image 266k no yes
codex/gpt-5.3-codex text+image 266k no yes
codex/gpt-5.4 text+image 266k no yes
codex/gpt-5.4-mini text+image 266k no yes
Current model status
"defaultModel": "codex/gpt-5.5",
"resolvedDefault": "codex/gpt-5.5"
but:
"allowed": [
"openai/gpt-5.5",
"openai/gpt-5.3-codex"
]
The active default model is not present in the allowed-model list.
Why This Appears To Be A Bug
The following simultaneously hold true:
This suggests a problem in model discovery, catalog normalization, provider mapping, or model-selection UI flow rather than a user configuration error.
Failure mode has been persistent for 2 instances on separate infrastructure across openclaw v26.5.22 / 5.26 / 5.27 / 5.28 (only release versions tested, maybe earlier versions affected)
Expected behavior
After successful authentication, openclaw config should display the previously working model picker.
Actual behavior
After successful authentication, openclaw config falls back to the main menu without ever displaying the previously working model picker.
OpenClaw version
2026.5.28
Operating system
Debian 13
Install method
npm
Model
codex plan / GPT-5.5 and GPT-5.3-codex tested
Provider / routing chain
openclaw-codex
Additional provider/model setup details
No response
Logs, screenshots, and evidence
Impact and severity
GPT-5.5 remains usable, but the model picker cannot be invoked
Additional information
This issue has been observed across multiple releases and was not resolved by upgrading to 2026.5.28.
The behavior appears related to the interaction between:
openai/*
codex/*
openai-codex
model/provider routing.
However, runtime inference itself appears healthy.
tested settings manually:
openclaw config set agents.defaults.model.primary openai/gpt-5.5
and
openclaw config set agents.defaults.model.primary codex/gpt-5.5
with the same missing-picker behavior.