## What happened?

When `silent_model` is configured on a deployment, the Chat Completions primary request succeeds, but Responses and Anthropic Messages primary requests return HTTP 500:

```text AsyncCompletions.create() got an unexpected keyword argument 'silent_model' ```

This is more severe than #31888, where the primary request succeeds but the mirrored request is not sent. In this case, the primary request also fails.

## Minimal reproduction

```yaml model_list: - model_name: primary-model litellm_params: model: openai/primary-model custom_llm_provider: <openai-compatible-provider> silent_model: shadow-model

- model_name: shadow-model litellm_params: model: openai/shadow-model custom_llm_provider: <openai-compatible-provider> ```

```bash curl http://localhost:4000/v1/responses \ -H "Authorization: Bearer $LITELLM_API_KEY" \ -H "Content-Type: application/json" \ -d '{"model":"primary-model","input":"Reply only OK"}' ```

Without `silent_model`, the Responses request returns HTTP 200. After adding only `silent_model`, the same request returns HTTP 500.

Calling `/v1/messages` with `silent_model` configured fails with the same…