fix(vertex): respect vertex_count_tokens_location for Claude count_tokens

The count_tokens handler unconditionally overrode vertex_location to
us-central1 for Claude models, ignoring the user-configured
vertex_count_tokens_location parameter. Also, us-central1 is no longer
a supported region — Google now supports us-east5, europe-west1, and
asia-southeast1.

Now vertex_count_tokens_location takes precedence, vertex_location is
used as fallback, and us-east5 is the default only when neither is set.

Fixes #23872

litellm/llms/vertex_ai/vertex_ai_partner_models/count_tokens/handler.py

@@ -105,12 +105,16 @@ class VertexAIPartnerModelsTokenCounter(VertexBase):
         # Extract Vertex AI credentials and settings
         vertex_credentials = self.get_vertex_ai_credentials(litellm_params)
         vertex_project = self.get_vertex_ai_project(litellm_params)
-        vertex_location = self.get_vertex_ai_location(litellm_params)
+        vertex_location = (
+            litellm_params.get("vertex_count_tokens_location")
+            or self.get_vertex_ai_location(litellm_params)
+        )
 
-        # Map empty location/cluade models to a supported region for count-tokens endpoint
+        # Default Claude models to us-east5 for count-tokens endpoint when no location is set
+        # Supported regions: us-east5, europe-west1, asia-southeast1
         # https://docs.cloud.google.com/vertex-ai/generative-ai/docs/partner-models/claude/count-tokens
-        if not vertex_location or "claude" in model.lower():
-            vertex_location = "us-central1"
+        if not vertex_location and "claude" in model.lower():
+            vertex_location = "us-east5"
 
         # Get access token and resolved project ID
         access_token, project_id = await self._ensure_access_token_async(

tests/test_litellm/llms/vertex_ai/vertex_ai_partner_models/count_tokens/test_count_tokens_location.py

@@ -0,0 +1,164 @@
+"""
+Tests for Vertex AI partner models count_tokens location resolution.
+
+Ref: https://github.com/BerriAI/litellm/issues/23872
+"""
+import pytest
+
+from litellm.llms.vertex_ai.vertex_ai_partner_models.count_tokens.handler import (
+    VertexAIPartnerModelsTokenCounter,
+)
+
+
+@pytest.fixture
+def counter():
+    return VertexAIPartnerModelsTokenCounter()
+
+
+class TestCountTokensLocationResolution:
+    """Verify that vertex_count_tokens_location is respected in handle_count_tokens_request."""
+
+    def _build_litellm_params(
+        self,
+        vertex_location=None,
+        vertex_count_tokens_location=None,
+    ):
+        params = {}
+        if vertex_location is not None:
+            params["vertex_location"] = vertex_location
+        if vertex_count_tokens_location is not None:
+            params["vertex_count_tokens_location"] = vertex_count_tokens_location
+        return params
+
+    @pytest.mark.asyncio
+    async def test_count_tokens_location_overrides_vertex_location(self, counter, monkeypatch):
+        """vertex_count_tokens_location should take precedence over vertex_location."""
+        captured = {}
+
+        async def fake_ensure_access_token(self, credentials, project_id, custom_llm_provider):
+            return "fake-token", "fake-project"
+
+        def fake_build_endpoint(self, model, project_id, vertex_location, api_base=None):
+            captured["vertex_location"] = vertex_location
+            return "https://fake-endpoint"
+
+        monkeypatch.setattr(
+            VertexAIPartnerModelsTokenCounter, "_ensure_access_token_async", fake_ensure_access_token
+        )
+        monkeypatch.setattr(
+            VertexAIPartnerModelsTokenCounter, "_build_count_tokens_endpoint", fake_build_endpoint
+        )
+
+        # Mock the HTTP call to avoid real network requests
+        class FakeResponse:
+            status_code = 200
+            def json(self):
+                return {"input_tokens": 10}
+            def raise_for_status(self):
+                pass
+
+        class FakeClient:
+            async def post(self, url, headers=None, json=None, **kwargs):
+                return FakeResponse()
+
+        import litellm.llms.vertex_ai.vertex_ai_partner_models.count_tokens.handler as handler_mod
+        monkeypatch.setattr(handler_mod, "get_async_httpx_client", lambda **kwargs: FakeClient())
+
+        litellm_params = self._build_litellm_params(
+            vertex_location="us-east5",
+            vertex_count_tokens_location="europe-west1",
+        )
+
+        await counter.handle_count_tokens_request(
+            model="claude-sonnet-4-6",
+            request_data={"messages": [{"role": "user", "content": "hi"}]},
+            litellm_params=litellm_params,
+        )
+
+        assert captured["vertex_location"] == "europe-west1"
+
+    @pytest.mark.asyncio
+    async def test_claude_without_count_tokens_location_defaults_to_us_east5(self, counter, monkeypatch):
+        """Claude models without any location should default to us-east5."""
+        captured = {}
+
+        async def fake_ensure_access_token(self, credentials, project_id, custom_llm_provider):
+            return "fake-token", "fake-project"
+
+        def fake_build_endpoint(self, model, project_id, vertex_location, api_base=None):
+            captured["vertex_location"] = vertex_location
+            return "https://fake-endpoint"
+
+        monkeypatch.setattr(
+            VertexAIPartnerModelsTokenCounter, "_ensure_access_token_async", fake_ensure_access_token
+        )
+        monkeypatch.setattr(
+            VertexAIPartnerModelsTokenCounter, "_build_count_tokens_endpoint", fake_build_endpoint
+        )
+
+        class FakeResponse:
+            status_code = 200
+            def json(self):
+                return {"input_tokens": 10}
+            def raise_for_status(self):
+                pass
+
+        class FakeClient:
+            async def post(self, url, headers=None, json=None, **kwargs):
+                return FakeResponse()
+
+        import litellm.llms.vertex_ai.vertex_ai_partner_models.count_tokens.handler as handler_mod
+        monkeypatch.setattr(handler_mod, "get_async_httpx_client", lambda **kwargs: FakeClient())
+
+        litellm_params = self._build_litellm_params()  # no location at all
+
+        await counter.handle_count_tokens_request(
+            model="claude-sonnet-4-6",
+            request_data={"messages": [{"role": "user", "content": "hi"}]},
+            litellm_params=litellm_params,
+        )
+
+        assert captured["vertex_location"] == "us-east5"
+
+    @pytest.mark.asyncio
+    async def test_claude_with_vertex_location_uses_it(self, counter, monkeypatch):
+        """Claude models with vertex_location but no count_tokens_location should use vertex_location."""
+        captured = {}
+
+        async def fake_ensure_access_token(self, credentials, project_id, custom_llm_provider):
+            return "fake-token", "fake-project"
+
+        def fake_build_endpoint(self, model, project_id, vertex_location, api_base=None):
+            captured["vertex_location"] = vertex_location
+            return "https://fake-endpoint"
+
+        monkeypatch.setattr(
+            VertexAIPartnerModelsTokenCounter, "_ensure_access_token_async", fake_ensure_access_token
+        )
+        monkeypatch.setattr(
+            VertexAIPartnerModelsTokenCounter, "_build_count_tokens_endpoint", fake_build_endpoint
+        )
+
+        class FakeResponse:
+            status_code = 200
+            def json(self):
+                return {"input_tokens": 10}
+            def raise_for_status(self):
+                pass
+
+        class FakeClient:
+            async def post(self, url, headers=None, json=None, **kwargs):
+                return FakeResponse()
+
+        import litellm.llms.vertex_ai.vertex_ai_partner_models.count_tokens.handler as handler_mod
+        monkeypatch.setattr(handler_mod, "get_async_httpx_client", lambda **kwargs: FakeClient())
+
+        litellm_params = self._build_litellm_params(vertex_location="asia-southeast1")
+
+        await counter.handle_count_tokens_request(
+            model="claude-sonnet-4-6",
+            request_data={"messages": [{"role": "user", "content": "hi"}]},
+            litellm_params=litellm_params,
+        )
+
+        assert captured["vertex_location"] == "asia-southeast1"