### Check for existing issues

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

### What happened?

### What happened?

With `gemini-3.5-flash`, a chat completion using `reasoning_effort` returns `completion_tokens_details.reasoning_tokens > 0`, but `choices[0].message.reasoning_content` is `None` and `thinking_blocks` is `[]`.

The identical setup with **`gemini-2.5-flash` works** and returns `reasoning_content` correctly. So this is specific to the 3.5 model family.

Passing `extra_body={"generationConfig": {"thinking_summaries": "auto"}}` (the new Gemini 3.5 summary switch) has **no effect** — response is still `reasoning_content=None`.

### Likely cause

Gemini changed how thought summaries are surfaced between 2.5 and 3.5: - **2.5**: summaries come back through the `generateContent` API via `includeThoughts` (which LiteLLM sets from `reasoning_effort`) → works. - **3.5**: `includeThoughts` is deprecated; summaries are controlled by `generationConfig.thinking_summaries: "auto"` and, per Google's docs, the `generateContent` API "has no dedicated thought blocks" — summaries are surfaced via the Interactions API. LiteLLM uses `genera…