From 955f6d8cae80d0199ec08ea34c318e2aa5108717 Mon Sep 17 00:00:00 2001
From: Arjun Pakhan <arjunpakhan@gmail.com>
Date: Mon, 14 Sep 2026 10:13:35 +0000
Subject: [PATCH 1/2] fix(anthropic): propagate and format custom api_base in
 count_tokens endpoint

---
 litellm/llms/anthropic/count_tokens/handler.py | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)

diff --git a/litellm/llms/anthropic/count_tokens/handler.py b/litellm/llms/anthropic/count_tokens/handler.py
index 38cd429d99a..9cdb886e5cd 100644
--- a/litellm/llms/anthropic/count_tokens/handler.py
+++ b/litellm/llms/anthropic/count_tokens/handler.py
@@ -66,8 +66,18 @@ class AnthropicCountTokensHandler(AnthropicCountTokensConfig):
 
             verbose_logger.debug("Transformed request: %s", request_body)
 
-            # Get endpoint URL
-            endpoint_url: Final = api_base or self.get_anthropic_count_tokens_endpoint()
+            # Construct clean endpoint URL handling custom proxy destinations
+            if api_base:
+                if api_base.endswith("/messages/count_tokens"):
+                    endpoint_url = api_base
+                elif api_base.endswith("/v1/messages"):
+                    endpoint_url = f"{api_base.rstrip('/')}/count_tokens"
+                elif api_base.endswith("/v1"):
+                    endpoint_url = f"{api_base.rstrip('/')}/messages/count_tokens"
+                else:
+                    endpoint_url = f"{api_base.rstrip('/')}/v1/messages/count_tokens"
+            else:
+                endpoint_url = self.get_anthropic_count_tokens_endpoint()
 
             verbose_logger.debug("Making request to: %s", endpoint_url)
 

From 0569cbe8d868d0ce369be4c0fa6d684815cd4f1c Mon Sep 17 00:00:00 2001
From: Arjun Pakhan <arjunpakhan@gmail.com>
Date: Mon, 14 Sep 2026 10:28:55 +0000
Subject: [PATCH 2/2] test(anthropic): add unit test coverage for count_tokens
 custom api_base endpoint resolution

---
 tests/test_anthropic_count_tokens.py | 32 ++++++++++++++++++++++++++++
 1 file changed, 32 insertions(+)
 create mode 100644 tests/test_anthropic_count_tokens.py

diff --git a/tests/test_anthropic_count_tokens.py b/tests/test_anthropic_count_tokens.py
new file mode 100644
index 00000000000..c14a0feb685
--- /dev/null
+++ b/tests/test_anthropic_count_tokens.py
@@ -0,0 +1,32 @@
+from unittest.mock import AsyncMock, patch
+
+import pytest
+
+from litellm.llms.anthropic.count_tokens.handler import AnthropicCountTokensHandler
+
+
+@pytest.mark.asyncio
+async def test_anthropic_count_tokens_custom_api_base():
+    handler = AnthropicCountTokensHandler()
+
+    mock_response = AsyncMock()
+    mock_response.status_code = 200
+    mock_response.json.return_value = {"input_tokens": 10}
+
+    with patch("litellm.llms.anthropic.count_tokens.handler.get_async_httpx_client") as mock_client_factory:
+        mock_client = AsyncMock()
+        mock_client.post.return_value = mock_response
+        mock_client_factory.return_value = mock_client
+
+        # Test custom api_base URL resolution and formatting
+        res = await handler.handle_count_tokens_request(
+            model="claude-3-5-sonnet-20241022",
+            messages=[{"role": "user", "content": "hello"}],
+            api_key="test-key",
+            api_base="https://custom-proxy.internal.com/v1",
+        )
+
+        assert res["input_tokens"] == 10
+        mock_client.post.assert_called_once()
+        called_url = mock_client.post.call_args[0][0]
+        assert called_url == "https://custom-proxy.internal.com/v1/messages/count_tokens"