#32683 [Bug]: Bedrock count_tokens returns 400 for cross-region inference-profile models — prefix stripped before request
### Check for existing issues
- [x] I have searched the existing issues and checked that my issue is not a duplicate.
### What happened?
POST /v1/messages/count_tokens returns 400 Bad Request for every Claude model configured behind an AWS Bedrock cross-region inference profile (region: eu-central-1). Chat/Converse completions to the same models work fine — only the token-counting endpoint fails.
Log excerpts:
HTTP error in CountTokens handler: Client error '400 Bad Request' for url 'https://bedrock-runtime.eu-central-1.amazonaws.com/model/anthropic.claude-opus-4-8/count-tokens' HTTP error in CountTokens handler: Client error '400 Bad Request' for url 'https://bedrock-runtime.eu-central-1.amazonaws.com/model/anthropic.claude-haiku-4-5-20251001-v1:0/count-tokens' HTTP error in CountTokens handler: Client error '400 Bad Request' for url 'https://bedrock-runtime.eu-central-1.amazonaws.com/model/anthropic.claude-opus-4-6-v1/count-tokens' HTTP error in CountTokens handler: Client error '400 Bad Request' for url 'https://bedrock-runtime.eu-central-1.amazonaws.com/model/anthropic.claude-sonnet-4-6/count-tokens' Root cause (traced through LiteLLM source, confirmed on v1.83.14)
/v1/me…