Broken vision for Anthropic models #1555

Describe the bug

I can't get any of the Anthropic Opus 4.5, Opus 4.6 or Sonnet 4.6 to read/understand images when using copilot-cli.

Either the request gets a timeout with:

or it responds back with something completely random about the image.

Example image attached, clearly a cat isn't it?! like copilot thinks it is when using opus/sonnet (when it's not timing out).

Using a different model, like gpt5.3-codex with copilot works perfectly fine.
Also, using claude code and the same anthropic models works perfectly fine, so the models are not the issue.

Affected version

GitHub Copilot CLI 0.0.412.

Steps to reproduce the behavior

1. Try sending it an image and ask it what is inside it
2. Either get a timeout back with an error that the model don't support vision, or a completely random wrong reply.

Expected behavior

Vision support and that it can actually see what's in the images.

Additional context

• Tried this on both windows and linux