### Check for existing issues

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

### What happened?

Gemini pass-through streaming endpoint (streamGenerateContent) returns truncated JSON. This issue seems intermittent when response contains large thoughtSignature.

When using the LiteLLM proxy's Gemini pass-through endpoint (/v1beta/models/{model}:streamGenerateContent), streaming responses from Gemini 3.x models with thinking enabled are intermittently truncated, resulting in invalid/incomplete JSON being delivered to the client.

This affects both google-genai SDK and the gemini-cli.

### Steps to Reproduce

1. LiteLLM version: [litellm[proxy]==1.89.0] 2. Models: Gemini Model with Vertex AI (global region) 3. Condition: Thinking/reasoning enabled (any thinkingLevel: minimal, low, medium, high)

### Relevant log output

```shell google.genai.errors.UnknownApiResponseError: Failed to parse response as JSON. ```

### What part of LiteLLM is this about?

Proxy

### What LiteLLM version are you on ?

v1.89.0

### Twitter / LinkedIn details

_No response_