Repository navigation
Restore last model choice used in session when it is resumed #14540
Description
What version of Codex CLI is running?
codex-cli 0.114.0
What subscription do you have?
Pro
Which model were you using?
Doesn't matter
What platform is your computer?
Win 11 x64
What terminal emulator and version are you using (if applicable)?
Windows Terminal, PS7.x
What issue are you seeing?
When changing the model or reasoning level in one session, resuming other sessions use that choice too instead of persisting with what they were using. Naturally I did not notice this until 5.4 xhigh kirby'd a week's worth of tokens in a day (usage tracking issues aside)
What steps can reproduce the bug?
- Use one session for a while set to a given model/reasoning level
- Exit codex
- In another session, change to a higher or lower amount of reasoning, or a different model
- Resume the other session
- Resumed session is now using what was chosen in the other without any indication of change
What is the expected behavior?
It doesn't do this, or at least calls it out. The difference between high and xhigh is not a sufficient amount of indication
Additional information
“He wins his battles by making no mistakes.” - Sun Tzu
Activity
github-actions commentedon Mar 13, 2026
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
etraut-openai commentedon Mar 13, 2026
The current behavior is by design, so I don't think we'd consider this a bug. Users previously gave us feedback that they wanted model choice to be persisted as the default. When resuming a session, the same default is then restored.
We don't currently store model or reasoning choices in the session's metadata since those can change throughout a session.
I'll change this from a bug report to a feature request, and we'll see what other users think.
[-]Model choice persisting unexpectedly into other sessions[/-][+]Restore last model choice used in session when it is resumed[/+]doublemover commentedon Mar 13, 2026
Wild. I've been using this for like 12 weeks and only just now noticed this. I suspect it may be a source of usage drain complaints!
A more visible notification would be lovely, along with the ability to opt out of changes in other sessions impacting this one on resume (by some sort of launch flag similar to
--last, and/or an option)etraut-openai commentedon Oct 12, 2026
(response from codex)
Thanks for the report. Resume behavior has changed since the original discussion. A normal resume now restores the thread's saved model and reasoning effort instead of taking those values from another session's defaults (#14888; included in CLI 0.162.1). Explicit model or reasoning overrides still take precedence when you supply them. We're closing this older report; please file a new issue if a current CLI loses the saved settings on a normal resume.