fix(vertex): bill Live API output cost when input_cost_per_token is zero

litellm/proxy/pass_through_endpoints/llm_provider_handlers/vertex_ai_live_passthrough_logging_handler.py

@@ -151,7 +151,9 @@ class VertexAILivePassthroughLoggingHandler(BasePassthroughLoggingHandler):
             verbose_proxy_logger.debug("Vertex AI Live API model info for '%s': %s", model, model_info)
 
             # Check if pricing info is available
-            if not model_info or not model_info.get("input_cost_per_token"):
+            if not model_info or (
+                not model_info.get("input_cost_per_token") and not model_info.get("output_cost_per_token")
+            ):
                 verbose_proxy_logger.error("No pricing info found for %s in local model pricing database", model)
                 return 0.0
 

tests/pass_through_unit_tests/test_vertex_ai_live_passthrough.py

@@ -227,6 +227,27 @@ class TestVertexAILivePassthroughLoggingHandler:
         assert cost >= expected_min_cost
         assert cost > 0
 
+    @patch(
+        "litellm.proxy.pass_through_endpoints.llm_provider_handlers.vertex_ai_live_passthrough_logging_handler.get_model_info"
+    )
+    def test_calculate_cost_input_free_output_priced(self, mock_get_model_info, handler):
+        """A model free on input but priced on output must still bill output, not return 0."""
+        mock_get_model_info.return_value = {
+            "input_cost_per_token": 0.0,
+            "output_cost_per_token": 0.000005,
+        }
+
+        usage_metadata = {
+            "promptTokenCount": 1000,
+            "candidatesTokenCount": 200,
+            "totalTokenCount": 1200,
+        }
+
+        cost = handler._calculate_live_api_cost("gemini-live-x", usage_metadata)
+
+        assert cost == pytest.approx(200 * 0.000005)
+        assert cost > 0
+
     @patch(
         "litellm.proxy.pass_through_endpoints.llm_provider_handlers.vertex_ai_live_passthrough_logging_handler.get_model_info"
     )