[BUG] API Provider "VS Code LM API" is getting a "Request Failed: 400" for all Language Model

[miniCristend]

Problem (one or two sentences)

I tried changing the model in the Language Model list, but they all return the following response: Request Failed: 400 {"error":{"message":"Model is not supported for this request.","code":"model_not_supported","param":"model","type":"invalid_request_error"}}

I also saw the following troubleshooting instruction, but no "Accept" window appeared: Note: Models accessed via the VS Code Language Model API may be wrapped or fine-tuned by the provider, so behavior can differ from using the same model directly from a typical provider or router. To use a model from the Language Model dropdown, first switch to that model and then click Accept in the Copilot Chat prompt; otherwise you may see an error such as 400 "The requested model is not supported".

Context (who is affected and when)

All models return 'Request Failed: 400' when prompted with any content.

Reproduction steps

1. Settings -> API Provider -> VS Code LM API -> Language Model -> any -> Save -> Done
2. New Task -> any mode -> any prompt -> Request Failed: 400

Expected result

The task starts and shows progress

Actual result

API Request Failed Request Failed: 400 {"error":{"message":"Model is not supported for this request.","code":"model_not_supported","param":"model","type":"invalid_request_error"}}

Variations tried (optional)

No response

App Version

Version: 3.29.5 (251afb3)

API Provider (optional)

VS Code Language Model API

Model Used (optional)

Any

Roo Code Task Links (optional)

No response

Relevant logs or errors (optional)

[roomote]

I'm exploring a potential fix for the VS Code LM API 400 error issue and will open a PR to share progress.

[roomote]

Opened a PR with my attempt at addressing this issue: #8992

The fix updates the VS Code LM API integration to use model IDs directly instead of vendor/family pairs, which was causing the 400 error. Also added better error messages to guide users when model approval is needed.

Feedback welcome!

[dvirhams]

Any workaround possible on the client side until issue is fixed?

[feng-java]

Is anyone working on this bug？

[johnib]

This appears to be the same issue as reported in the Cline repository. See the root cause analysis here: cline#7229 (comment)

The issue is that the GitHub Copilot API has started blocking requests that include the x-onbehalf-extension-id header. The VS Code LM API adds this header automatically to indicate the requesting extension, but GitHub Copilot's API is now rejecting these requests with a "model not supported" error.

When GitHub Copilot makes requests directly, the header isn't included and they succeed. However, when using the VS Code LM API as an intermediary (which both Cline and Roo-Code do), the header gets added, triggering the API rejection.

[k00bas]

just to add: downgrading the versions of roo-code, gitbub copilot or github copilot chat extensions hasn't help; same for cleaning the VS Code cache, so I believe currently there's no workaround other than waiting for the fix on copilot API side