Merge 61e4266138 into f2118fc767

litellm/llms/vertex_ai/vertex_llm_base.py

@@ -112,19 +112,8 @@ class VertexBase:
         supported_regions: Final = model_info.get("supported_regions")
 
         if supported_regions and len(supported_regions) > 0:
-            # If user didn't specify region, use the first supported region
             if vertex_region is None:
                 return supported_regions[0]
-            # If user specified a region not supported by this model, override it
-            if vertex_region not in supported_regions:
-                verbose_logger.warning(
-                    "Vertex AI model '%s' does not support region '%s' (supported: %s). Routing to '%s'.",
-                    model,
-                    vertex_region,
-                    supported_regions,
-                    supported_regions[0],
-                )
-                return supported_regions[0]
             return vertex_region
         return vertex_region or "us-central1"
 

litellm/model_prices_and_context_window_backup.json

@@ -30216,6 +30216,11 @@
         "supported_output_modalities": [
             "text"
         ],
+        "supported_regions": [
+            "global",
+            "us",
+            "eu"
+        ],
         "supports_audio_input": true,
         "supports_function_calling": true,
         "supports_parallel_function_calling": true,
@@ -30284,6 +30289,11 @@
         "supported_output_modalities": [
             "text"
         ],
+        "supported_regions": [
+            "global",
+            "us",
+            "eu"
+        ],
         "supports_audio_input": true,
         "supports_function_calling": true,
         "supports_minimal_reasoning_effort": false,
@@ -76358,6 +76368,11 @@
         "supported_output_modalities": [
             "text"
         ],
+        "supported_regions": [
+            "global",
+            "us",
+            "eu"
+        ],
         "supports_audio_input": true,
         "supports_function_calling": true,
         "supports_minimal_reasoning_effort": false,

model_prices_and_context_window.json

@@ -30216,6 +30216,11 @@
         "supported_output_modalities": [
             "text"
         ],
+        "supported_regions": [
+            "global",
+            "us",
+            "eu"
+        ],
         "supports_audio_input": true,
         "supports_function_calling": true,
         "supports_parallel_function_calling": true,
@@ -30284,6 +30289,11 @@
         "supported_output_modalities": [
             "text"
         ],
+        "supported_regions": [
+            "global",
+            "us",
+            "eu"
+        ],
         "supports_audio_input": true,
         "supports_function_calling": true,
         "supports_minimal_reasoning_effort": false,
@@ -76358,6 +76368,11 @@
         "supported_output_modalities": [
             "text"
         ],
+        "supported_regions": [
+            "global",
+            "us",
+            "eu"
+        ],
         "supports_audio_input": true,
         "supports_function_calling": true,
         "supports_minimal_reasoning_effort": false,

tests/unit/llms/vertex_ai/test_vertex_ai_common_utils.py

@@ -1069,10 +1069,8 @@ def test_get_vertex_url_global_region(stream, expected_endpoint_suffix):
     [
         # Model with supported_regions=["global"], no user region -> use "global"
         ({"supported_regions": ["global"]}, None, "global"),
-        # Model with supported_regions=["global"], user passes unsupported region -> override to "global"
-        ({"supported_regions": ["global"]}, "us-central1", "global"),
-        # Model with supported_regions=["global"], user passes unsupported region -> override to "global"
-        ({"supported_regions": ["global"]}, "europe-west1", "global"),
+        ({"supported_regions": ["global"]}, "us-central1", "us-central1"),
+        ({"supported_regions": ["global"]}, "europe-west1", "europe-west1"),
         # Model with supported_regions=["us-west2"], no user region -> use "us-west2"
         ({"supported_regions": ["us-west2"]}, None, "us-west2"),
         # Model with supported_regions=["us-west2", "us-central1"], user passes supported region -> respect it
@@ -1081,11 +1079,10 @@ def test_get_vertex_url_global_region(stream, expected_endpoint_suffix):
             "us-central1",
             "us-central1",
         ),
-        # Model with supported_regions=["us-west2", "us-central1"], user passes unsupported region -> override
         (
             {"supported_regions": ["us-west2", "us-central1"]},
             "europe-west1",
-            "us-west2",
+            "europe-west1",
         ),
         # No model_cost entry, no user region -> default us-central1
         ({}, None, "us-central1"),
@@ -1112,6 +1109,32 @@ def test_get_vertex_region_global_only_model(model_cost_entry, vertex_region, ex
         assert result == expected_region
 
 
+@pytest.mark.parametrize("model", ["gemini-3.7-flash", "gemini-3.8-flash", "gemini-4-argon"])
+@pytest.mark.parametrize(
+    "vertex_region, expected_region",
+    [
+        (None, "global"),
+        ("global", "global"),
+        ("us", "us"),
+        ("eu", "eu"),
+        ("us-central1", "us-central1"),
+        ("europe-west1", "europe-west1"),
+    ],
+)
+def test_gemini_defaults_to_global_and_preserves_explicit_regions(
+    model: str, vertex_region: str | None, expected_region: str
+) -> None:
+    """Location availability verified 2026-10-19:
+
+    https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/3-7-flash
+    https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/3-8-flash
+    https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/4-argon
+    """
+    from litellm.llms.vertex_ai.vertex_llm_base import VertexBase
+
+    assert VertexBase.get_vertex_region(vertex_region=vertex_region, model=model) == expected_region
+
+
 def test_vertex_filter_format_uri():
     import json
 

tests/unit/llms/vertex_ai/vertex_ai_partner_models/gemma/test_vertex_ai_gemma_global_endpoint.py

@@ -105,7 +105,7 @@ class TestVertexBaseGetVertexRegionGemma:
             )
             assert result == "global"
 
-    def test_global_model_with_unsupported_user_region_overrides(self):
+    def test_global_model_preserves_explicit_user_region(self):
         vertex_base = VertexBase()
 
         with patch.dict(
@@ -121,7 +121,7 @@ class TestVertexBaseGetVertexRegionGemma:
                 vertex_region="us-central1",
                 model="google/gemma-4-26b-a4b-it-maas",
             )
-            assert result == "global"
+            assert result == "us-central1"
 
 
 class TestCreateVertexURLGemma:

tests/unit/llms/vertex_ai/vertex_ai_partner_models/qwen/test_vertex_ai_qwen_global_endpoint.py

@@ -64,8 +64,8 @@ class TestVertexBaseGetVertexRegion:
             )
             assert result == "global"
 
-    def test_global_model_with_unsupported_user_region_overrides(self):
-        """Test that unsupported user region is overridden for global-only models."""
+    def test_global_model_preserves_explicit_user_region(self):
+        """Test that an explicit user region is not silently broadened."""
         vertex_base = VertexBase()
 
         with patch.dict(
@@ -81,7 +81,7 @@ class TestVertexBaseGetVertexRegion:
                 vertex_region="us-central1",
                 model="qwen/qwen3-next-80b-a3b-instruct-maas",
             )
-            assert result == "global"
+            assert result == "us-central1"
 
     def test_non_global_model_uses_provided_region(self):
         """Test that non-global models use the provided region."""