fix(proxy): compare model_info echo against the deployment's cost-map lookup

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

litellm/proxy/management_endpoints/model_management_endpoints.py

@@ -138,7 +138,6 @@ from litellm.types.router import (
     updateLiteLLMParams,
 )
 from litellm.types.utils import (
-    COST_MAP_LOOKUP_KEY,
     echoed_cost_map_fields,
     echoed_cost_map_pricing_fields,
     without_server_derived_pricing,
@@ -876,12 +875,11 @@ def _ptu_priced_deployment(model_params: Deployment) -> Deployment:
     )
 
 
-def _cost_map_entry(model_info: Mapping[str, object]) -> Mapping[str, object]:
-    key: Final = model_info.get(COST_MAP_LOOKUP_KEY)
-    if not isinstance(key, str):
-        return MappingProxyType({})
+def _cost_map_entry(db_model: Deployment, incoming_model_info: Mapping[str, object]) -> Mapping[str, object]:
+    base_model: Final = incoming_model_info.get("base_model")
+    lookup: Final = base_model if isinstance(base_model, str) else db_model.litellm_params.model
     try:
-        return MappingProxyType(dict(litellm.get_model_info(model=key)))
+        return MappingProxyType(dict(litellm.get_model_info(model=lookup)))
     except Exception:
         return MappingProxyType({})
 
@@ -909,7 +907,9 @@ def update_db_model(db_model: Deployment, updated_patch: updateDeployment) -> Pr
     # update model info
     if updated_patch.model_info:
         incoming_model_info: Final = updated_patch.model_info.model_dump(exclude_none=True)
-        echoed_fields: Final = echoed_cost_map_fields(incoming_model_info, _cost_map_entry(incoming_model_info))
+        echoed_fields: Final = echoed_cost_map_fields(
+            incoming_model_info, _cost_map_entry(db_model, incoming_model_info)
+        )
         merged_model_info.update(
             MappingProxyType(
                 dict(

tests/test_litellm/proxy/management_endpoints/test_model_management_endpoints.py

@@ -3961,7 +3961,7 @@ class TestModelInfoCostMapEchoFilter:
         from litellm.proxy.management_endpoints.model_management_endpoints import update_db_model
         from litellm.types.router import Deployment, LiteLLM_Params, ModelInfo
 
-        entry = litellm.get_model_info("gpt-5.6")
+        entry = litellm.get_model_info("openai/gpt-5.6")
         echo = {**entry, "id": "dep-echo-0", "db_model": True, "access_groups": ["prod"]}
         db_model = Deployment(
             model_name="gpt-5.6",
@@ -3987,7 +3987,7 @@ class TestModelInfoCostMapEchoFilter:
         from litellm.proxy.management_endpoints.model_management_endpoints import update_db_model
         from litellm.types.router import Deployment, LiteLLM_Params, ModelInfo
 
-        entry = litellm.get_model_info("gpt-5.6")
+        entry = litellm.get_model_info("openai/gpt-5.6")
         echo = {
             **entry,
             "id": "dep-echo-1",
@@ -4020,7 +4020,7 @@ class TestModelInfoCostMapEchoFilter:
         from litellm.types.router import Deployment, LiteLLM_Params, ModelInfo
         from litellm.types.utils import echoed_cost_map_fields
 
-        entry = litellm.get_model_info("gpt-5.6")
+        entry = litellm.get_model_info("openai/gpt-5.6")
         assert echoed_cost_map_fields({"max_input_tokens": entry["max_input_tokens"]}, entry) == ()
         db_model = Deployment(
             model_name="gpt-5.6",
@@ -4049,7 +4049,7 @@ class TestModelInfoCostMapEchoFilter:
         from litellm.proxy.management_endpoints.model_management_endpoints import update_db_model
         from litellm.types.router import Deployment, LiteLLM_Params, ModelInfo
 
-        entry = litellm.get_model_info("gpt-5.6")
+        entry = litellm.get_model_info("openai/gpt-5.6")
         db_model = Deployment(
             model_name="gpt-5.6",
             litellm_params=LiteLLM_Params(model="openai/gpt-5.6"),
@@ -4072,7 +4072,7 @@ class TestModelInfoCostMapEchoFilter:
         from litellm.proxy.management_endpoints.model_management_endpoints import update_db_model
         from litellm.types.router import Deployment, LiteLLM_Params, ModelInfo
 
-        entry = litellm.get_model_info("gpt-5.6")
+        entry = litellm.get_model_info("openai/gpt-5.6")
         db_model = Deployment(
             model_name="gpt-5.6",
             litellm_params=LiteLLM_Params(model="openai/gpt-5.6"),
@@ -4090,6 +4090,76 @@ class TestModelInfoCostMapEchoFilter:
         assert info["mode"] == "chat"
         assert info["access_groups"] == ["staging"]
 
+    def test_echo_is_compared_against_the_deployments_lookup_not_the_key(self):
+        import litellm
+
+        from litellm.proxy.management_endpoints.model_management_endpoints import update_db_model
+        from litellm.types.router import Deployment, LiteLLM_Params, ModelInfo
+
+        lookup_pairs: Final = (
+            ("openai/gpt-5.6", "gpt-5.6"),
+            ("openai/gpt-4.1-mini", "gpt-4.1-mini"),
+        )
+        lookup_data: Final = tuple(
+            (deployment_model, deployment_entry, differing_fields)
+            for deployment_model, key_model in lookup_pairs
+            for deployment_entry in (litellm.get_model_info(deployment_model),)
+            for key_entry in (litellm.get_model_info(key_model),)
+            for differing_fields in (
+                frozenset(
+                    k for k in deployment_entry if k in key_entry and deployment_entry[k] != key_entry[k]
+                ),
+            )
+            if differing_fields
+        )
+        if not lookup_data:
+            pytest.skip("No deployment/key cost-map lookup differences are available")
+
+        deployment_model, entry, differing_fields = lookup_data[0]
+        assert differing_fields
+        db_model = Deployment(
+            model_name=deployment_model,
+            litellm_params=LiteLLM_Params(model=deployment_model),
+            model_info=ModelInfo(id="dep-echo-5"),
+        )
+        echo = {**entry, "id": "dep-echo-5", "db_model": True, "access_groups": ["prod"]}
+
+        result = update_db_model(
+            db_model=db_model,
+            updated_patch=updateDeployment(model_info=ModelInfo(**echo)),
+        )
+
+        info = json.loads(result["model_info"])
+        assert not frozenset(info).intersection(frozenset(entry) - frozenset(("mode",)))
+
+    def test_base_model_wins_over_litellm_params_model_for_the_lookup(self):
+        import litellm
+
+        from litellm.proxy.management_endpoints.model_management_endpoints import update_db_model
+        from litellm.types.router import Deployment, LiteLLM_Params, ModelInfo
+
+        entry = litellm.get_model_info("azure/gpt-5.6")
+        db_model = Deployment(
+            model_name="azure/my-deploy",
+            litellm_params=LiteLLM_Params(model="azure/my-deploy"),
+            model_info=ModelInfo(id="dep-echo-6", base_model="azure/gpt-5.6"),
+        )
+        echo = {
+            **entry,
+            "id": "dep-echo-6",
+            "base_model": "azure/gpt-5.6",
+            "db_model": True,
+        }
+
+        result = update_db_model(
+            db_model=db_model,
+            updated_patch=updateDeployment(model_info=ModelInfo(**echo)),
+        )
+
+        info = json.loads(result["model_info"])
+        assert not frozenset(info).intersection(frozenset(entry) - frozenset(("mode",)))
+        assert info["base_model"] == "azure/gpt-5.6"
+
 
 class TestUpdateDBModelClearCacheControlInjectionPoints:
     def test_explicit_null_removes_stored_injection_points(self):