### Check for existing issues

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

### What happened?

With LiteLLM Proxy response caching (Redis) enabled and a reasoning model behind it:

- Cache MISS (streamed): client receives delta.reasoning_content chunks as expected - Cache entry in Redis: contains reasoning_content in the stored message - Cache HIT (streamed): the response contains NO reasoning_content. Clients silently lose the reasoning on every cache hit

Helmfile release:

``` - name: litellm namespace: litellm chart: berriai/litellm-helm version: 1.89.1 ```

Helm values.yaml (shortened):

``` image: # Official LiteLLM image with database support repository: ghcr.io/berriai/litellm-non_root tag: "1.89.1"

proxy_config: litellm_settings: request_timeout: 600 json_logs: true

cache: true cache_params: ttl: 172800 # 2 days type: redis host: os.environ/REDIS_HOST port: os.environ/REDIS_PORT password: os.environ/REDIS_PASSWORD ```

### Steps to Reproduce

1. Send streamed request twice:

``` curl -k -v https://LITELLM_HOST/v1/chat/completions \ -H "Authorization: Bearer…