Anthropic model selected via Foundry AI Toolkit in Copilot fails with 400 (max_tokens: Field required)

I’m seeing a reproducible failure when using a Foundry-deployed Anthropic model through VS Code Copilot model selection. (Works with non-Anthropic model in same setup).

Environment

• VS Code: macOS
• Extension: ms-windows-ai-studio.windows-ai-studio (Foundry AI Toolkit) v: 1.4.3
• Copilot Chat enabled

Steps to Reproduce

1. Install Foundry AI Toolkit in VS Code.
2. Sign in and connect to a Foundry project.
3. Open Copilot model manager and select a deployed Anthropic model from the connected Foundry project.
4. Send a normal chat prompt. (I sent 'test').

Actual Result
Copilot request fails with a 400 error from the inference path.

Error shown:

2026-06-16 14:53:11.923 [error] Unable to call the (<PROJECT>)claude-opus-4-7 inference endpoint due to 400. Please check if the input or configuration is correct. 400 {"type":"error","error":{"type":"invalid_request_error","message":"max_tokens: Field required"},"request_id":"<REQUEST_ID>"}

• invalid_request_error
• max_tokens: Field required
  (Observed in extension stack trace under ms-windows-ai-studio.windows-ai-studio.)

Expected Result
Request should succeed, or the extension should correctly map/send required Anthropic fields (including max_tokens) when invoking chat.

Notes

• Authentication and project connection are valid.
• The issue appears specific to Anthropic invocation through this extension flow.
• Works with OpenAI Models in same setup
• Works with other providers and models provided by CoPilot. Issues only stems from accessing models through foundry extension.