From ffd46eab5fefef8292accd7046bae4a1fdde6b44 Mon Sep 17 00:00:00 2001 From: Sameer Kankute Date: Fri, 1 May 2026 18:16:38 +0530 Subject: [PATCH] docs(gemini): document Gemini 3 thinking defaults and opt-in flag Made-with: Cursor --- docs/providers/gemini.md | 8 ++++++++ 1 file changed, 8 insertions(+) diff --git a/docs/providers/gemini.md b/docs/providers/gemini.md index dcc0f2f5e..6edb21a1b 100644 --- a/docs/providers/gemini.md +++ b/docs/providers/gemini.md @@ -295,7 +295,15 @@ curl http://0.0.0.0:4000/v1/chat/completions \ +For **Gemini 3+ models**, LiteLLM now follows provider defaults by default and does **not** force `thinkingLevel` when you pass Anthropic-style `thinking={"type":"enabled","budget_tokens":...}`. +If you want the legacy LiteLLM behavior (force `thinkingLevel="low"` for Pro and `thinkingLevel="minimal"` for Flash), enable: + +```python +import litellm + +litellm.enable_gemini_default_thinking_level_low = True +``` ## Usage - `service_tier`