What version of Codex CLI is running?
0.144.0
What subscription do you have?
ChatGPT authentication
Which model were you using?
gpt-5.6
What platform is your computer?
Ubuntu 24.04 LTS
What terminal emulator and version are you using (if applicable)?
No response
Codex doctor report
What issue are you seeing?
Summary
I'm seeing a platform-specific issue where GPT-5.6 models work on macOS but consistently fail on Linux, despite using:
- the same OpenAI account
- the same ChatGPT authentication
- the same reported Codex CLI version (
0.144.0)
macOS can use GPT-5.6 Sol/Terra/Luna normally.
Linux consistently returns:
Model metadata for 'gpt-5.6-sol' not found.
Defaulting to fallback metadata; this can degrade performance and cause issues.
followed by:
{
"type": "error",
"status": 400,
"error": {
"type": "invalid_request_error",
"message": "The 'gpt-5.6-sol' model requires a newer version of Codex. Please upgrade to the latest app or CLI and try again."
}
}
However, codex doctor reports that Linux is already running the latest available version.
Environment
Linux
- Ubuntu 24.04 LTS
- Codex CLI 0.144.0
- ChatGPT authentication
- Same OpenAI account as macOS
Output:
Output:
/home/user/.local/bin/codex
macOS
- macOS 15
- Codex CLI 0.144.0
- Same OpenAI account
Output:
GPT-5.6 works correctly on macOS.
Steps to reproduce
(or gpt-5.6-terra / gpt-5.6-luna)
Expected behavior
The GPT-5.6 model should start normally, as it does on macOS.
Actual behavior
Linux reports:
Model metadata for 'gpt-5.6-sol' not found.
then immediately returns:
400 invalid_request_error
The 'gpt-5.6-sol' model requires a newer version of Codex.
Please upgrade to the latest app or CLI and try again.
Troubleshooting already performed
Verified:
- Same OpenAI account
- Same ChatGPT authentication
- Same Codex CLI version (0.144.0)
- Fresh login (
codex logout / codex login)
- Reinstalled Codex
- Reinstalled via npm
kernel.unprivileged_userns_clone=1
- Bubblewrap installed and working
- Network connectivity OK
codex doctor reports the CLI is already up to date
Nothing above changes the behavior.
Interesting observation
One noticeable difference is the rollout information reported by codex doctor.
Linux
Linux also reports:
rollout files are missing from the state DB;
state DB rows point at missing or unusable rollout files
macOS
The macOS doctor reports a different rollout warning:
rollout scan was incomplete or found bad files
I don't know whether the rollout counts are expected to differ across platforms, but the difference is significant, and the Linux client is also the only one that cannot find GPT-5.6 model metadata.
This may or may not be related, but it seemed worth mentioning.
Additional information
I'm happy to provide:
- Full
codex doctor output
- Linux/macOS comparison
- Screenshots
- Any additional logs if helpful
Thanks!
What steps can reproduce the bug?
Reproduction steps
- Install the latest available Codex CLI on Linux (Ubuntu 24.04).
codex --version
# codex-cli 0.144.0
- Authenticate with ChatGPT.
- Launch any GPT-5.6 model.
(Also reproducible with gpt-5.6-terra and gpt-5.6-luna.)
- Enter any prompt, for example:
- The request immediately fails with:
Model metadata for 'gpt-5.6-sol' not found.
Defaulting to fallback metadata; this can degrade performance and cause issues.
followed by:
400 invalid_request_error
The 'gpt-5.6-sol' model requires a newer version of Codex.
Please upgrade to the latest app or CLI and try again.
Expected behavior
The model should start normally, as it does on macOS using the same OpenAI account and the same reported Codex CLI version (0.144.0).
Notes
- This reproduces consistently on Ubuntu 24.04.
- The same account can successfully use
gpt-5.6-sol on macOS with Codex CLI 0.144.0.
codex doctor reports that the Linux installation is already up to date.
What is the expected behavior?
No response
Additional information
No response
What version of Codex CLI is running?
0.144.0
What subscription do you have?
ChatGPT authentication
Which model were you using?
gpt-5.6
What platform is your computer?
Ubuntu 24.04 LTS
What terminal emulator and version are you using (if applicable)?
No response
Codex doctor report
What issue are you seeing?
Summary
I'm seeing a platform-specific issue where GPT-5.6 models work on macOS but consistently fail on Linux, despite using:
0.144.0)macOS can use GPT-5.6 Sol/Terra/Luna normally.
Linux consistently returns:
followed by:
{ "type": "error", "status": 400, "error": { "type": "invalid_request_error", "message": "The 'gpt-5.6-sol' model requires a newer version of Codex. Please upgrade to the latest app or CLI and try again." } }However,
codex doctorreports that Linux is already running the latest available version.Environment
Linux
Output:
Output:
macOS
Output:
GPT-5.6 works correctly on macOS.
Steps to reproduce
(or
gpt-5.6-terra/gpt-5.6-luna)Expected behavior
The GPT-5.6 model should start normally, as it does on macOS.
Actual behavior
Linux reports:
then immediately returns:
Troubleshooting already performed
Verified:
codex logout/codex login)kernel.unprivileged_userns_clone=1codex doctorreports the CLI is already up to dateNothing above changes the behavior.
Interesting observation
One noticeable difference is the rollout information reported by
codex doctor.Linux
Linux also reports:
macOS
The macOS doctor reports a different rollout warning:
I don't know whether the rollout counts are expected to differ across platforms, but the difference is significant, and the Linux client is also the only one that cannot find GPT-5.6 model metadata.
This may or may not be related, but it seemed worth mentioning.
Additional information
I'm happy to provide:
codex doctoroutputThanks!
What steps can reproduce the bug?
Reproduction steps
codex --version # codex-cli 0.144.0(Also reproducible with
gpt-5.6-terraandgpt-5.6-luna.)followed by:
Expected behavior
The model should start normally, as it does on macOS using the same OpenAI account and the same reported Codex CLI version (
0.144.0).Notes
gpt-5.6-solon macOS with Codex CLI0.144.0.codex doctorreports that the Linux installation is already up to date.What is the expected behavior?
No response
Additional information
No response