test: restore synthetic behavior tests dropped as catalog pins

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

tests/test_litellm/litellm_core_utils/test_fallback_generalizations.py

@@ -859,6 +859,27 @@ def test_router_registration_does_not_shadow_shipped_rules(shipped_cost_map):
     assert litellm.supports_reasoning(model="claude-opus-9", custom_llm_provider="anthropic") is True
 
 
+def test_deployment_model_info_beats_the_seeded_rule_defaults(shipped_cost_map):
+    """Seeding a registration from the rules is a floor, not an override: an explicit
+    model_info on the deployment still wins, so a non-reasoning model can be configured
+    under a reasoning-first namespace."""
+    from litellm import Router
+
+    model = "wandb/some-org/NoThink-1"
+    Router(
+        model_list=[
+            {
+                "model_name": model,
+                "litellm_params": {"model": model, "api_key": "fake"},
+                "model_info": {"supports_reasoning": False},
+            }
+        ]
+    )
+
+    assert litellm.model_cost[model]["supports_reasoning"] is False
+    assert litellm.supports_reasoning(model="some-org/NoThink-1", custom_llm_provider="wandb") is False
+
+
 def test_shipped_rules_flag_unmapped_openai_reasoning_families(shipped_cost_map):
     for model in (
         "gpt-5.7-nova",

tests/test_litellm/llms/bedrock/test_bedrock_common_utils.py

@@ -441,6 +441,34 @@ def test_explicit_invoke_route_does_not_match_async_invoke():
     )
 
 
+def test_capability_lookups_fall_back_to_base_model_when_regional_entry_lacks_field(monkeypatch):
+    """
+    Regression test: a regional model_cost entry without the capability field
+    must not shadow a base entry that has it (`get(model) or get(base)` used to
+    short-circuit on the truthy regional dict and drop the capability).
+    """
+    import litellm
+    from litellm.llms.bedrock.common_utils import (
+        bedrock_converse_supports_parallel_tool_use_config,
+        is_claude_4_5_on_bedrock,
+    )
+
+    base = "anthropic.claude-fallback-test"
+    regional = f"eu.{base}"
+    monkeypatch.setitem(litellm.model_cost, regional, {"input_cost_per_token": 1e-06})
+    monkeypatch.setitem(
+        litellm.model_cost,
+        base,
+        {
+            "cache_creation_input_token_cost_above_1hr": 1e-05,
+            "supports_parallel_tool_use_config": True,
+        },
+    )
+
+    assert is_claude_4_5_on_bedrock(regional) is True
+    assert bedrock_converse_supports_parallel_tool_use_config(regional) is True
+
+
 def test_merge_bedrock_aws_request_params_strips_caller_identity_when_deployment_has_static_credentials():
     from litellm.llms.bedrock.common_utils import merge_bedrock_aws_request_params
 

tests/test_litellm/llms/vertex_ai/test_vertex_ai_common_utils.py

@@ -143,6 +143,24 @@ def test_anyof_with_excessive_nesting():
         convert_anyof_null_to_nullable(schema)
 
 
+@pytest.mark.asyncio
+async def test_get_supports_system_message():
+    """Test get_supports_system_message with different models"""
+    from litellm.llms.vertex_ai.common_utils import get_supports_system_message
+
+    # fine-tuned vertex gemini models will specifiy they are in the /gemini spec format
+    result = get_supports_system_message(
+        model="gemini/1234567890", custom_llm_provider="vertex_ai"
+    )
+    assert result == True
+
+    # non-fine-tuned vertex gemini models will not specifiy they are in the /gemini spec format
+    result = get_supports_system_message(
+        model="random-model-name", custom_llm_provider="vertex_ai"
+    )
+    assert result == False
+
+
 @pytest.mark.parametrize(
     "model, expected",
     [

tests/test_litellm/llms/vertex_ai/vertex_ai_partner_models/anthropic/test_vertex_ai_partner_models_anthropic_messages_config.py

@@ -452,6 +452,44 @@ def test_vertex_claude_completion_does_not_mutate_shared_extra_headers():
     assert shared_extra_headers == {}, "extra_headers must not be mutated by completion()"
 
 
+def test_messages_thinking_shape_follows_exact_vertex_entry_flag(local_model_cost_map, monkeypatch):
+    """The Vertex messages config must probe capabilities under ``vertex_ai`` so an
+    operator setting ``supports_adaptive_thinking: false`` on the exact
+    ``vertex_ai/claude-opus-4-8`` entry beats the unmodified ``anthropic`` entry.
+    With the inherited ``"anthropic"`` provider default the flip was ignored and
+    the transform kept emitting ``thinking.type='adaptive'``."""
+    import litellm
+
+    config = VertexAIPartnerModelsAnthropicMessagesConfig()
+
+    def transform():
+        return config.transform_anthropic_messages_request(
+            model="claude-opus-4-8",
+            messages=[{"role": "user", "content": "Hello"}],
+            anthropic_messages_optional_request_params={
+                "max_tokens": 4096,
+                "reasoning_effort": "medium",
+            },
+            litellm_params=GenericLiteLLMParams(),
+            headers={},
+        )
+
+    result = transform()
+    assert result.get("thinking") == {"type": "adaptive", "display": "summarized"}
+    assert result.get("output_config") == {"effort": "medium"}
+
+    monkeypatch.setitem(litellm.model_cost["vertex_ai/claude-opus-4-8"], "supports_adaptive_thinking", False)
+    litellm.get_model_info.cache_clear()
+    assert litellm.model_cost["claude-opus-4-8"]["supports_adaptive_thinking"] is True
+
+    flipped = transform()
+    thinking = flipped.get("thinking")
+    assert isinstance(thinking, dict)
+    assert thinking.get("type") == "enabled"
+    assert isinstance(thinking.get("budget_tokens"), int)
+    assert "output_config" not in flipped
+
+
 def _vertex_transform(model, messages, system=None):
     config = VertexAIPartnerModelsAnthropicMessagesConfig()
     params = {"max_tokens": 256}