From 2303379c2080aaeca398a7a3d3b8f5ac2019cc60 Mon Sep 17 00:00:00 2001
From: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
Date: Sat, 19 Sep 2026 00:07:22 -0700
Subject: [PATCH] fix(gemini): keep the 2048 cache minimum on Gemini 2.5 Pro
 only, per Google's live cachedContents API

---
 litellm/model_prices_and_context_window_backup.json   |  2 --
 model_prices_and_context_window.json                  |  2 --
 .../context_caching/test_vertex_ai_context_caching.py | 11 ++++++-----
 3 files changed, 6 insertions(+), 9 deletions(-)

diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json
index fcdf6c4baa3..de387552a44 100644
--- a/litellm/model_prices_and_context_window_backup.json
+++ b/litellm/model_prices_and_context_window_backup.json
@@ -25066,7 +25066,6 @@
         "supports_parallel_function_calling": true,
         "supports_pdf_input": true,
         "supports_prompt_caching": true,
-        "prompt_cache_min_tokens": 2048,
         "supports_reasoning": true,
         "supports_response_schema": true,
         "supports_system_messages": true,
@@ -27139,7 +27138,6 @@
         "supports_parallel_function_calling": true,
         "supports_pdf_input": true,
         "supports_prompt_caching": true,
-        "prompt_cache_min_tokens": 2048,
         "supports_reasoning": true,
         "supports_response_schema": true,
         "supports_system_messages": true,
diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json
index fcdf6c4baa3..de387552a44 100644
--- a/model_prices_and_context_window.json
+++ b/model_prices_and_context_window.json
@@ -25066,7 +25066,6 @@
         "supports_parallel_function_calling": true,
         "supports_pdf_input": true,
         "supports_prompt_caching": true,
-        "prompt_cache_min_tokens": 2048,
         "supports_reasoning": true,
         "supports_response_schema": true,
         "supports_system_messages": true,
@@ -27139,7 +27138,6 @@
         "supports_parallel_function_calling": true,
         "supports_pdf_input": true,
         "supports_prompt_caching": true,
-        "prompt_cache_min_tokens": 2048,
         "supports_reasoning": true,
         "supports_response_schema": true,
         "supports_system_messages": true,
diff --git a/tests/test_litellm/llms/vertex_ai/context_caching/test_vertex_ai_context_caching.py b/tests/test_litellm/llms/vertex_ai/context_caching/test_vertex_ai_context_caching.py
index 5c33b9a995b..7cbfacfc338 100644
--- a/tests/test_litellm/llms/vertex_ai/context_caching/test_vertex_ai_context_caching.py
+++ b/tests/test_litellm/llms/vertex_ai/context_caching/test_vertex_ai_context_caching.py
@@ -1396,14 +1396,15 @@ class TestContextCachingEndpoints:
         # Restart the patcher so teardown_method can stop it cleanly
         self._token_check_patcher.start()
 
-    @pytest.mark.parametrize("model", ["gemini-2.5-flash", "gemini-2.5-pro"])
-    def test_check_and_create_cache_skips_between_default_and_gemini_2_5_minimum(
-        self, model, local_model_cost_map
+    def test_check_and_create_cache_skips_between_default_and_gemini_2_5_pro_minimum(
+        self, local_model_cost_map
     ):
-        """Gemini 2.5 Flash and Pro need 2048 cached tokens, twice the provider-agnostic default.
+        """Gemini 2.5 Pro needs 2048 cached tokens, twice the provider-agnostic default.
 
-        Content between the two used to reach Google's cachedContents endpoint and 400.
+        Content between the two used to reach Google's cachedContents endpoint and 400
+        with "Cached content is too small".
         """
+        model = "gemini-2.5-pro"
         self._token_check_patcher.stop()
 
         cached_messages = [