BYOK Azure Foundry /responses fails with invalid input type (input[n].type)

Summary

Using VS Code Copilot Chat BYOK with an Azure Foundry OpenAI-compatible endpoint and explicit /responses path can fail with a request validation error indicating an invalid/missing input[n].type.

Environment

• VS Code: latest stable (repro observed in May 2026)
• Extension: GitHub Copilot Chat (BYOK custom model)
• Provider in chatLanguageModels.json: azure
• Endpoint style: Azure Foundry project OpenAI-compatible endpoint

Config Snippet

{
  "models": {
    "provider": "azure",
    "url": "https://<foundry-host>.services.ai.azure.com/api/projects/<project>/openai/v1/responses",
    "id": "<model-id>",
    "name": "<display-name>",
    "toolCalling": true,
    "vision": false,
    "maxInputTokens": 64000,
    "maxOutputTokens": 8000
  }
}

Steps to Reproduce

1. Configure a BYOK Azure custom model in chatLanguageModels.json with an explicit /responses URL (Foundry project endpoint).
2. Select that model in Copilot Chat.
3. Send a simple prompt (for example: Hello).
4. Observe request failure.

Actual Result

Request fails with 400-like validation response (sanitized):

{
  "error": {
    "message": "Invalid value: ''. Supported values are: 'input_text', 'input_image', 'output_text', 'refusal', 'input_file', 'computer_screenshot', and 'summary_text'.",
    "type": "invalid_request_error",
    "param": "input[1]",
    "code": "invalid_value"
  }
}

Expected Result

The BYOK request payload generated for /responses should be compatible with Azure Foundry OpenAI-compatible Responses API expectations.

Notes

• Using /chat/completions is more reliable in this setup.
• This may be a payload-shape compatibility issue in the BYOK Responses path for Azure Foundry endpoints.