### Check for existing issues

- [x] I have searched the existing issues and checked that my issue is not a duplicate.

### What happened?

Setup: Claude Code → Anthropic protocol → litellm proxy → OpenAI protocol → glm-5.1

What happens: Model returns correct function.arguments (e.g. '{"file_path": "test.py"}'), but after litellm converts the response to Anthropic format, the tool_use.input becomes {} (empty dict).

Evidence from logs:

OpenAI response (correct): arguments='{"file_path": "D:\\python-code\\test\\app.py"}' Anthropic response sent to client (broken): input: {} Next request from Claude Code includes the broken input: {} Regression: v1.79.1 does NOT have this issue. v1.83.7 does.

Related code changes in v1.83.7:

New _remove_thought_signatures_from_messages function with model_dump() calls New _process_assistant_message_tool_calls with model_dump() on tool_call Pydantic models from __future__ import annotations added to utils.py

### Steps to Reproduce

1. 2. 3.

### Relevant log output

```shell

```

### What part of LiteLLM is this about?

_No response_

### What LiteLLM version are you on ?

v1.83.7-stable

### Twitter / LinkedIn details

_No response_