Repository navigation
BYOK statusline shows medium effort in 1.0.41 despite --effort high #3135
Description
Description
After upgrading to Copilot CLI 1.0.41, when using --effort high with a BYOK/custom provider, the actual request is still sent with reasoning_effort: "high", but the statusline receives model.display_name as gpt-5.5 (medium).
The same setup works correctly in Copilot CLI 1.0.40, where the statusline displays gpt-5.5 (high).
This looks like a regression in the statusline/UI display state: the actual request effort and the effort shown through the UI/statusline are out of sync.
Environment
- Copilot CLI: 1.0.41
- OS: macOS
- Provider: BYOK / custom OpenAI-compatible provider
- Wire API:
responses - Model:
gpt-5.5
Relevant environment variables:
COPILOT_PROVIDER_BASE_URL=http://<custom-provider>/v1
COPILOT_PROVIDER_API_KEY=<redacted>
COPILOT_PROVIDER_WIRE_API=responses
COPILOT_PROVIDER_TYPE=openai
COPILOT_MODEL=gpt-5.5
COPILOT_PROVIDER_MODEL_ID=gpt-5.5
COPILOT_PROVIDER_WIRE_MODEL=gpt-5.5Launch command:
copilot --effort highExpected behavior
The statusline payload should reflect the reasoning effort that is actually being used.
For example, with --effort high, the statusline payload should contain:
{
"model": {
"id": "gpt-5.5",
"display_name": "gpt-5.5 (high)"
}
}Actual behavior
In 1.0.41, the statusline displays:
gpt-5.5 (medium)
That means the statusline payload display name is effectively:
{
"model": {
"id": "gpt-5.5",
"display_name": "gpt-5.5 (medium)"
}
}However, the debug log shows that the actual request is using high:
Using custom provider: type=openai, baseUrl=http://<custom-provider>/v1, wireApi=responses
"model": "gpt-5.5"
"defaultReasoningEffort": "high"
"reasoning_effort": "high"
So the issue appears to be limited to the UI/statusline display state, not the actual request.
Comparison test
Using the same environment, same BYOK provider, same model, and same --effort high flag, only changing the Copilot CLI version:
| Version | Actual request reasoning_effort |
Statusline model.display_name |
|---|---|---|
| 1.0.40 | high |
gpt-5.5 (high) |
| 1.0.41 | high |
gpt-5.5 (medium) |
Captured statusline payload from 1.0.40:
{
"model": {
"id": "gpt-5.5",
"display_name": "gpt-5.5 (high)"
}
}Captured request settings from the 1.0.41 debug log:
"defaultReasoningEffort": "high"
"reasoning_effort": "high"
But the 1.0.41 statusline shows:
gpt-5.5 (medium)
Activity
Mapleeeeeeeeeee commentedon May 7, 2026
Fixed in v1.0.43. Thanks!
Mapleeeeeeeeeee commentedon May 7, 2026
Additional repro details after testing locally:
Environment:
Launch:
Observed on v1.0.43:
gpt-5.5 · medium.gpt-5.5-eu2 (gpt-5.5) (medium).From inspecting the packaged app.js in the CLI binary, the footer/model hook appears to apply the CLI reasoning effort only once, then later the BYOK/custom-provider branch refreshes the UI state using the model default:
The provider default resolver ultimately falls back to
defaultReasoningEffort ?? "medium". Forgpt-5.5, there is no explicitdefaultReasoningEffort, so the footer display falls back to medium even when the request path still sends high.This makes the built-in footer/statusline unreliable for BYOK: with
--effort high, the API receives high but the footer displays medium.sschyu commentedon Aug 6, 2026
Similar issue here. The status line for the built-in GPT model also shows "Medium", though the effort settings are
high/xhigh.