You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[Bug]: No enabled canonical OpenAI provider for model: gpt-5.6-sol #261
After installing OpenCodex and starting it with ocx start, requests made from the Codex app using a GPT model fail with the following error:
unexpected status 404 Not Found: No enabled canonical OpenAI provider for model: gpt-5.6-sol, url: http://127.0.0.1:10100/v1/responses
OpenCodex version:
Reproduction
ocx start
Open the Codex app.
Select a GPT model.
Send any request.
The request fails with a 404 Not Found error:
unexpected status 404 Not Found: No enabled canonical OpenAI provider for model: gpt-5.6-sol, url: http://127.0.0.1:10100/v1/responses
Logs and screenshots
Area
CLI
Version
2.7.31
OS
macOS 26.5.2 Mac mini M4
Config shape
Checks
I searched existing issues and docs first.
I removed secrets, tokens, and personal data from logs and screenshots.
This appears to be a configuration issue rather than a bug.
The error "No enabled canonical OpenAI provider for model: gpt-5.6-sol" means the proxy has no provider configured to route requests to. When you install OpenCodex and run ocx start without first running ocx init, there's no config file with a provider, so the proxy can't route anything.
To fix:
Run ocx init to configure a provider (e.g., OpenAI, Anthropic, etc.)
Then run ocx start
The default config that ships with OpenCodex includes an OpenAI provider that works out of the box with Codex's ChatGPT OAuth, but if you ran ocx init and selected a different provider, or if your config got corrupted, you'll need to reconfigure.
I've also improved the error message in the next release to be more actionable: "No enabled OpenAI provider for model: gpt-5.6-sol. Run 'ocx init' to configure a provider, or check that your config has an enabled 'openai' provider."
I’m really sorry. I just realized that I had set Kimi as the default provider.
After reinstalling OpenCodex and running ocx init with OpenAI selected as the default provider, the issue was resolved. Since the problem is now fixed, I’ll close this issue.
However, I think the wording shown during ocx init could be improved:
🔧 opencodex (ocx) setup
Available providers:
This wording made it seem as though I was selecting all providers or models that should be made available, rather than choosing the default provider to use.
I’d like to suggest changing the prompt to something clearer, such as:
Activity
Wibias commentedon Jul 22, 2026
This appears to be a configuration issue rather than a bug.
The error "No enabled canonical OpenAI provider for model: gpt-5.6-sol" means the proxy has no provider configured to route requests to. When you install OpenCodex and run ocx start without first running ocx init, there's no config file with a provider, so the proxy can't route anything.
To fix:
The default config that ships with OpenCodex includes an OpenAI provider that works out of the box with Codex's ChatGPT OAuth, but if you ran ocx init and selected a different provider, or if your config got corrupted, you'll need to reconfigure.
I've also improved the error message in the next release to be more actionable: "No enabled OpenAI provider for model: gpt-5.6-sol. Run 'ocx init' to configure a provider, or check that your config has an enabled 'openai' provider."
raydian commentedon Jul 22, 2026
That is not the issue. I executed "ocx init", selected a third-party model, ran "ocx start", restarted Codex, and the issue still occurs.
detail config command:
#236 (comment)
#236 (comment)
#236 (comment)
Wibias commentedon Jul 22, 2026
Is this your second account? If not, lets not mix things up and open your own issue, if that still persists after it got closed.
bloooi commentedon Jul 22, 2026
I’m really sorry. I just realized that I had set Kimi as the default provider.
After reinstalling OpenCodex and running
ocx initwith OpenAI selected as the default provider, the issue was resolved. Since the problem is now fixed, I’ll close this issue.However, I think the wording shown during
ocx initcould be improved:This wording made it seem as though I was selecting all providers or models that should be made available, rather than choosing the default provider to use.
I’d like to suggest changing the prompt to something clearer, such as:
Thank you, and sorry again for the confusion.
fix(router): improve error message when no OpenAI provider is configured
fix(router): improve error message when no OpenAI provider is configured
fix(cli): clarify ocx init prompts to indicate default provider selec…