[Bug]: 'invalid beta flag' for Claude models on VertexAI

[steve-gore-snapdocs]

What happened?

With Roo Code configured to use the LiteLLM provider with claude-sonnet-4-5 provided by VertexAI and prompt caching enabled, we get the error:

litellm.BadRequestError: Vertex_aiException BadRequestError - b'{"type":"error","error":{"type":"invalid_request_error","message":"invalid beta flag"},"request_id":"req_vrtx_xxxxxxx"}'. Received Model Group=claude-sonnet-4-5
Available Model Group Fallbacks=None LiteLLM Retried: 1 times, LiteLLM Max Retries: 2

With prompt caching disabled, the request returns successfully.

I believe this was broken in https://github.com/BerriAI/litellm/pull/17142/changes#diff-3e49c88c9e4541d46e9443e79f9b34c48a9ff0e29768fbbcc417018866c41f1b. There was a fix made for Bedrock (https://github.com/BerriAI/litellm/pull/17301/changes) but not VertexAI. Additionally, some of the logic is different between the Bedrock and Vertex implementation (Vertex does not start with provided headers, etc).

@Sameerlite

Relevant log output

What part of LiteLLM is this about?

Proxy

What LiteLLM version are you on ?

v1.80.8

Twitter / LinkedIn details

No response

[Sameerlite]

@steve-gore-snapdocs we solved this here few days back: #18037
Does this not solve your issue? Can you try with latest code on main?

[cege7480]

I've found that I can have it enabled if I have a custom temperature as 1

[genesis-gh-agorshkov]

Also, how come the fix is for Vertex only?

anthropic-beta headers were deprecated, I believe they should be removed everywhere, not just vertex:

anthropics/claude-cookbooks#175