From 8828f002bea2d9842fb95fae78c90d60bc7ce41d Mon Sep 17 00:00:00 2001
From: Chesars <cesarponce19544@gmail.com>
Date: Tue, 17 Mar 2026 23:21:24 -0300
Subject: [PATCH] fix(gemini): pass model to context caching URL builder for
 custom api_base

_get_token_and_url_context_caching() was hardcoding model=None when
calling _check_custom_proxy(), which raises ValueError when api_base
is set because Gemini proxy URLs need the model name:
{api_base}/models/{model}:cachedContents

Fixes #23846
---
 .../vertex_ai_context_caching.py              |  5 ++-
 .../test_vertex_ai_context_caching.py         | 39 ++++++++++++++++++-
 2 files changed, 42 insertions(+), 2 deletions(-)

diff --git a/litellm/llms/vertex_ai/context_caching/vertex_ai_context_caching.py b/litellm/llms/vertex_ai/context_caching/vertex_ai_context_caching.py
index db6be9499a..c2e064d656 100644
--- a/litellm/llms/vertex_ai/context_caching/vertex_ai_context_caching.py
+++ b/litellm/llms/vertex_ai/context_caching/vertex_ai_context_caching.py
@@ -51,6 +51,7 @@ class ContextCachingEndpoints(VertexBase):
         vertex_project: Optional[str],
         vertex_location: Optional[str],
         vertex_auth_header: Optional[str],
+        model: Optional[str] = None,
     ) -> Tuple[Optional[str], str]:
         """
         Internal function. Returns the token and url for the call.
@@ -89,7 +90,7 @@ class ContextCachingEndpoints(VertexBase):
             stream=None,
             auth_header=auth_header,
             url=url,
-            model=None,
+            model=model,
             vertex_project=vertex_project,
             vertex_location=vertex_location,
             vertex_api_version="v1beta1"
@@ -342,6 +343,7 @@ class ContextCachingEndpoints(VertexBase):
             vertex_project=vertex_project,
             vertex_location=vertex_location,
             vertex_auth_header=vertex_auth_header,
+            model=model,
         )
 
         headers = {
@@ -488,6 +490,7 @@ class ContextCachingEndpoints(VertexBase):
             vertex_project=vertex_project,
             vertex_location=vertex_location,
             vertex_auth_header=vertex_auth_header,
+            model=model,
         )
 
         headers = {
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 3f8cbf1236..11ccd34804 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
@@ -1317,4 +1317,41 @@ class TestVertexAIGlobalLocation:
             # Assert correct URL format for global with beta API
             expected_url = "https://aiplatform.googleapis.com/v1beta1/projects/test-project/locations/global/cachedContents"
             assert url == expected_url, f"Expected {expected_url}, got {url}"
-            assert "global-aiplatform" not in url, "URL should not contain 'global-aiplatform' prefix"
\ No newline at end of file
+            assert "global-aiplatform" not in url, "URL should not contain 'global-aiplatform' prefix"
+
+    def test_gemini_context_caching_with_custom_api_base_passes_model(self):
+        """Gemini context caching with custom api_base must pass model to _check_custom_proxy.
+
+        Regression test for https://github.com/BerriAI/litellm/issues/23846
+        Previously model was hardcoded to None, causing ValueError when api_base was set.
+        """
+        caching = ContextCachingEndpoints()
+
+        auth_header, url = caching._get_token_and_url_context_caching(
+            gemini_api_key="test-key",
+            custom_llm_provider="gemini",
+            api_base="https://my-proxy.example.com",
+            vertex_project=None,
+            vertex_location=None,
+            vertex_auth_header=None,
+            model="gemini-1.5-pro",
+        )
+
+        assert "models/gemini-1.5-pro" in url
+        assert url.startswith("https://my-proxy.example.com/")
+
+    def test_gemini_context_caching_without_api_base_ignores_model(self):
+        """Without custom api_base, model param is not needed (default URL is used)."""
+        caching = ContextCachingEndpoints()
+
+        auth_header, url = caching._get_token_and_url_context_caching(
+            gemini_api_key="test-key",
+            custom_llm_provider="gemini",
+            api_base=None,
+            vertex_project=None,
+            vertex_location=None,
+            vertex_auth_header=None,
+        )
+
+        assert "generativelanguage.googleapis.com" in url
+        assert "cachedContents" in url
\ No newline at end of file