Bug summary

Selecting Raptor mini (Preview) in VS Code Copilot Chat fails consistently with HTTP 400.

The response includes:

• invalid_request_body
• Invalid type for 'context_management': expected an object, but got an array instead.

Repro steps

1. Open VS Code Insiders with GitHub Copilot Chat enabled.
2. In Copilot Chat model picker, select Raptor mini (Preview).
3. Send any prompt in default chat panel.
4. Observe failure.

Actual result

Chat fails with:

• Copilot Request id: 7886a781-5676-4255-a5c7-ea6e0c0aa44f
• GH Request Id: E0BC:BB827:612A1E:7538A8:69A25607
• Error: Request Failed: 400 {"error":{"message":"Invalid type for 'context_management': expected an object, but got an array instead.","code":"invalid_request_body"}}

Expected result

Raptor mini requests should succeed in the same context where other Copilot models succeed.

Evidence from local Copilot logs

From:
~/.vscode-server-insiders/data/logs/20260227T230410/exthost2/GitHub.copilot-chat/GitHub Copilot Chat.log

Relevant lines:

• 2026-02-27 23:42:16.454 [info] Request ID for failed request: 7886a781-5676-4255-a5c7-ea6e0c0aa44f
• 2026-02-27 23:42:16.458 [error] Server error: 400 {"error":{"message":"Invalid type for 'context_management': expected an object, but got an array instead.","code":"invalid_request_body"}}
• 2026-02-27 23:42:16.458 [error] Request Failed: 400 {"error":{"message":"Invalid type for 'context_management': expected an object, but got an array instead.","code":"invalid_request_body"}}
• 2026-02-27 23:42:16.464 [info] ccreq:48972e7f.copilotmd | failed | oswe-vscode-prime | 2236ms | [panel/editAgent]

Same session also shows successful requests with non-Raptor models (e.g. gpt-5.3-codex), suggesting issue is specific to this model/payload path.

Environment

• VS Code: 1.110.0-insider
• VS Code quality: insider
• Commit: 37f39645643438347f1b5077abdaa96e7297fa7c
• GitHub Copilot Chat extension: 0.38.2026022704
• OS: Linux 6.6.87.2-microsoft-standard-WSL2 x86_64
• Account: GitHub Copilot Pro

Additional context

The model is visible and selectable in the model picker (Raptor mini (Preview)), but every invocation returns the same 400 payload validation error.