gpt-5.5 on ChatGPT account is routed to Responses-Lite and fails with unsupported_value
Summary
On Windows Codex / codex-cli using a ChatGPT subscription account (auth_mode = "chatgpt") and the built-in openai-http provider (https://chatgpt.com/backend-api/codex), gpt-5.5 consistently fails with:
This model is not supported when using X-OpenAI-Internal-Codex-Responses-Lite.
gpt-5.4 and gpt-5.4-mini still work on the same machine and same account.
Environment
- Windows 11
- Codex Desktop + codex-cli
- codex-cli versions tested:
0.142.3,0.142.4 auth_mode = "chatgpt"model_provider = "openai-http"base_url = "https://chatgpt.com/backend-api/codex"wire_api = "responses"- ChatGPT subscription account (no API key)
Repro
Run a minimal request:
codex exec -m gpt-5.5 --skip-git-repo-check --dangerously-bypass-approvals-and-sandbox "Reply with exactly: OK"
Actual result
Fails with:
{
"error": {
"message": "This model is not supported when using X-OpenAI-Internal-Codex-Responses-Lite.",
"type": "invalid_request_error",
"param": "model",
"code": "unsupported_value"
}
}
Control checks
gpt-5.4worksgpt-5.4-miniworks- Re-login via
codex logout+codex logincompleted successfully, issue persists - Bypassing the local VPN/proxy does not change the result
- This is not limited to a long conversation/thread; minimal fresh requests reproduce it
- Same account on macOS can use
gpt-5.5
Additional clue
The local model catalog/cache marks gpt-5.5 with use_responses_lite = true, while gpt-5.4 and gpt-5.4-mini are false. The failure strongly suggests that gpt-5.5 is being routed into a Responses-Lite path that no longer accepts it.
Expected result
gpt-5.5 should either:
- not be routed through Responses-Lite for ChatGPT account Codex traffic, or
- be rejected earlier in model selection/UI with a clear compatibility message instead of failing at request time.
Please check whether the ChatGPT-account Codex route is incorrectly forcing gpt-5.5 into X-OpenAI-Internal-Codex-Responses-Lite.
This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗