Auto model fallback on quota limit shows new model in status line but does not resume — requires full restart
- Dominant language
- Shell
- Stars
- 11.2k
- Forks
- 1.9k
- Avg merge
- 14h 16m
- Merged PRs (30d)
- 6
Description
### Describe the bug
When the Copilot premium request quota is exhausted mid-session and the model is set to **Auto** (to automatically fall back to another model when the limit is hit), the CLI detects the switch and updates the bottom-left status line to reflect the new model (e.g., `Auto -> GPT-4o`), but **does not resume** the pending response or allow the session to continue. No further prompts are processed. The only recovery is to fully exit and relaunch the CLI.
### Affected version
GitHub Copilot CLI 1.0.44.
### Steps to reproduce the behavior
1. Open Copilot CLI (`copilot`).
2. Use `/model` to set the model to **Auto** (automatic fallback when quota is reached).
3. Submit prompts until the premium request quota is exhausted.
4. Observe the bottom-left status line — it updates to show the fallback model (e.g., `Auto → GPT-4o`).
5. Attempt to continue the conversation or re-submit the last prompt. It will still tell you you've reached limit.
6. Notice the CLI is stuck, no response is generated and no new prompts are processed.
### Expected behavior
After the model automatically switches to the Auto-selected fallback, the CLI should **seamlessly resume**, either retrying the last request automatically or allowing the user to continue the session without any interruption.
### Additional context
- **Operating system:** Linux
- **CPU architecture:** x86_64
- **Terminal emulator:** GNOME Terminal
- **Shell:** bash
- **Installation method:** curl install script
- **Experimental mode enabled:** no
- **Model at time of quota hit:** Claude Sonnet 4.6 (current default)
- **Fallback model shown in status line:** `Auto → GPT-4o`
**Workaround:** Exiting (`ctrl+d` or `ctrl+c×2`) and relaunching `copilot` starts a new session that uses the Auto-selected fallback model normally.
**Note:** The status line correctly reflects the new model after the switch, suggesting the fallback model selection logic *does* trigger, the bug is in the resumption step after switching.
Contributor guide
Assessment
This issue has not been assessed yet.