fix(proxy): drop echoed cost-map pricing on a row's next save and build /model/info pricing stamps without mutation

litellm/proxy/management_endpoints/model_management_endpoints.py

@@ -137,7 +137,7 @@ from litellm.types.router import (
     updateDeployment,
     updateLiteLLMParams,
 )
-from litellm.types.utils import without_server_derived_pricing
+from litellm.types.utils import echoed_cost_map_pricing_fields, without_server_derived_pricing
 from litellm.utils import get_utc_datetime
 
 if TYPE_CHECKING:
@@ -876,7 +876,11 @@ def update_db_model(db_model: Deployment, updated_patch: updateDeployment) -> Pr
         _raise_if_ptu_cost_attribution_disabled(updated_patch.model_info.model_dump(exclude_none=True))
     merged_model_name: Final = updated_patch.model_name or db_model.model_name
     merged_litellm_params: Final = db_model.litellm_params.model_dump(exclude_none=True)
-    merged_model_info: Final[dict[str, object]] = db_model.model_info.model_dump(exclude_none=True)
+    stored_model_info: Final = db_model.model_info.model_dump(exclude_none=True)
+    echoed_pricing: Final = echoed_cost_map_pricing_fields(stored_model_info)
+    merged_model_info: Final[dict[str, object]] = {
+        k: v for k, v in stored_model_info.items() if k not in echoed_pricing
+    }
 
     # update litellm params
     if updated_patch.litellm_params:

litellm/proxy/proxy_server.py

@@ -13630,12 +13630,17 @@ def _enrich_model_info_with_litellm_data(
     discovered_model_info: Final = (
         llm_router.get_discovered_model_info(model_info.get("id")) if llm_router is not None else MappingProxyType({})
     )
-    for k, v in _pricing_override_stamps(model_info, model.get("litellm_params") or MappingProxyType({})).items():
-        model_info[k] = v
-    for k, v in MappingProxyType({**litellm_model_info, **discovered_model_info}).items():
-        if k not in model_info or (model_info[k] is None and k in discovered_model_info):
-            model_info[k] = v
-    model["model_info"] = model_info
+    stamped_model_info: Final = MappingProxyType(
+        {**model_info, **_pricing_override_stamps(model_info, model.get("litellm_params") or MappingProxyType({}))}
+    )
+    model["model_info"] = {
+        **stamped_model_info,
+        **{
+            k: v
+            for k, v in MappingProxyType({**litellm_model_info, **discovered_model_info}).items()
+            if k not in stamped_model_info or (stamped_model_info[k] is None and k in discovered_model_info)
+        },
+    }
     # don't return the api key / vertex credentials
     # don't return the llm credentials
     model = remove_sensitive_info_from_deployment(model, excluded_keys={"litellm_credential_name"})
@@ -15101,12 +15106,17 @@ def _get_proxy_model_info(model: dict) -> dict:
     discovered_model_info: Final = (
         llm_router.get_discovered_model_info(model_info.get("id")) if llm_router is not None else MappingProxyType({})
     )
-    for k, v in _pricing_override_stamps(model_info, model.get("litellm_params") or MappingProxyType({})).items():
-        model_info[k] = v
-    for k, v in MappingProxyType({**litellm_model_info, **discovered_model_info}).items():
-        if k not in model_info or (model_info[k] is None and k in discovered_model_info):
-            model_info[k] = v
-    model["model_info"] = model_info
+    stamped_model_info: Final = MappingProxyType(
+        {**model_info, **_pricing_override_stamps(model_info, model.get("litellm_params") or MappingProxyType({}))}
+    )
+    model["model_info"] = {
+        **stamped_model_info,
+        **{
+            k: v
+            for k, v in MappingProxyType({**litellm_model_info, **discovered_model_info}).items()
+            if k not in stamped_model_info or (stamped_model_info[k] is None and k in discovered_model_info)
+        },
+    }
     # don't return the llm credentials
     model = remove_sensitive_info_from_deployment(deployment_dict=model, excluded_keys={"litellm_credential_name"})
 

tests/test_litellm/proxy/management_endpoints/test_model_management_endpoints.py

@@ -3734,6 +3734,66 @@ class TestModelInfoServerDerivedPricingFilter:
         assert info["access_groups"] == ["prod"]
         assert "pricing_overrides" not in info
 
+    def test_a_row_pinned_before_1_102_drops_its_cost_map_copy_on_its_next_save(self, monkeypatch: pytest.MonkeyPatch):
+        """LIT-8064. A stored ``model_info`` carrying ``key`` is a ``/model/info`` response an old
+        UI wrote back, so its pricing is the cost map of that day. The next edit of the row, here
+        only its reasoning level, leaves that copy behind and keeps everything the operator set."""
+        from litellm.proxy.common_utils.encrypt_decrypt_utils import decrypt_value_helper
+        from litellm.proxy.management_endpoints.model_management_endpoints import (
+            update_db_model,
+        )
+        from litellm.types.router import Deployment, LiteLLM_Params, ModelInfo
+
+        monkeypatch.setenv("LITELLM_SALT_KEY", "sk-lit8064-heal-on-save")
+        db_model = Deployment(
+            model_name="gpt-5.6",
+            litellm_params=LiteLLM_Params(model="openai/gpt-5.6", reasoning_effort="medium"),
+            model_info=ModelInfo(
+                id="dep-pinned-0",
+                key="gpt-5.6",
+                mode="chat",
+                access_groups=["prod"],
+                input_cost_per_token=4e-06,
+                output_cost_per_token=2e-05,
+                cache_read_input_token_cost_above_272k_tokens=8e-07,
+            ),
+        )
+
+        result = update_db_model(
+            db_model=db_model,
+            updated_patch=updateDeployment(litellm_params=updateLiteLLMParams(reasoning_effort="low")),
+        )
+
+        info = json.loads(result["model_info"])
+        params = json.loads(result["litellm_params"])
+        assert decrypt_value_helper(value=params["reasoning_effort"], key="reasoning_effort") == "low"
+        assert (info["key"], info["mode"], info["access_groups"]) == ("gpt-5.6", "chat", ["prod"])
+        for field in ("input_cost_per_token", "output_cost_per_token", "cache_read_input_token_cost_above_272k_tokens"):
+            assert field not in info, f"{field} still pins the row to the cost map of the day it was saved"
+            assert field not in params
+
+    def test_a_litellm_params_price_survives_the_cost_map_copy_being_dropped(self):
+        """The price an operator typed on ``litellm_params`` is the override the customer asked
+        for, so dropping the echoed ``model_info`` copy must leave it in place."""
+        from litellm.proxy.management_endpoints.model_management_endpoints import (
+            update_db_model,
+        )
+        from litellm.types.router import Deployment, LiteLLM_Params, ModelInfo
+
+        db_model = Deployment(
+            model_name="gpt-5.6",
+            litellm_params=LiteLLM_Params(model="openai/gpt-5.6", input_cost_per_token=3e-06),
+            model_info=ModelInfo(id="dep-typed-0", key="gpt-5.6", input_cost_per_token=3e-06),
+        )
+
+        result = update_db_model(
+            db_model=db_model,
+            updated_patch=updateDeployment(model_info=ModelInfo(id="dep-typed-0", access_groups=["prod"])),
+        )
+
+        assert json.loads(result["litellm_params"])["input_cost_per_token"] == 3e-06
+        assert json.loads(result["model_info"])["access_groups"] == ["prod"]
+
     def test_tiered_above_threshold_pricing_is_dropped(self):
         """Tiered rates ride `get_model_info` on a pattern match and are declared on no
         model, so a filter built only from the declared pricing fields would miss them."""

tests/test_litellm/proxy/proxy_server/test_proxy_config.py

@@ -2709,6 +2709,37 @@ def test_ProxyConfig__add_deployment_pinned_row_follows_the_cost_map_across_relo
     assert litellm.model_cost["typed-row"]["input_cost_per_token"] == 3e-06
 
 
+def test_ProxyConfig__add_deployment_ptu_row_with_a_cost_map_copy_still_bills_zero(monkeypatch, local_model_cost_map):
+    """A PTU deployment bills nothing per token: the proxy writes zeros to both blobs. When such
+    a row also carries the echoed cost map, dropping the ``model_info`` copy must not send it
+    back to the per-token price, because the ``litellm_params`` zeros are the operator's."""
+    monkeypatch.setattr(
+        "litellm.proxy.proxy_server.decrypt_value_helper",
+        lambda value, key, return_original_value: value,
+    )
+    router = litellm.Router(model_list=[])
+    monkeypatch.setattr("litellm.proxy.proxy_server.llm_router", router)
+    ptu = SimpleNamespace(
+        model_id="ptu-row",
+        model_name="gpt-5.6-ptu",
+        model_info={**PINNED_MODEL_INFO, "id": "ptu-row", "input_cost_per_token": 0.0, "output_cost_per_token": 0.0},
+        litellm_params={
+            "model": "openai/gpt-5.6",
+            "api_key": "sk-test",
+            "input_cost_per_token": 0.0,
+            "output_cost_per_token": 0.0,
+        },
+        blocked=False,
+    )
+
+    assert ProxyConfig()._add_deployment(db_models=[ptu]) == 1
+    router._replay_model_cost_registrations()
+
+    assert litellm.model_cost["ptu-row"]["input_cost_per_token"] == 0.0
+    assert litellm.model_cost["ptu-row"]["output_cost_per_token"] == 0.0
+    assert router.get_deployment(model_id="ptu-row").model_info.input_cost_per_token == 0.0
+
+
 def test_ProxyConfig_get_model_info_with_id_missing_model_id_raises(monkeypatch):
     monkeypatch.setattr("litellm.proxy.proxy_server.premium_user", False)
     pc = ProxyConfig()

tests/test_litellm/proxy/proxy_server/test_routes_model_info.py

@@ -328,6 +328,49 @@ def test_get_proxy_model_info_names_config_model_info_pricing_as_an_override(mon
     assert info["output_cost_per_token"] == 7e-06
 
 
+def test_v2_model_info_reports_pricing_overrides_to_the_admin_ui(client, auth_as, monkeypatch, local_model_cost_map):
+    """LIT-8064. The Admin UI model page reads ``GET /v2/model/info``, so the override report
+    has to ride that route too, not only ``/model/info``."""
+    model_list: Final = [
+        {
+            "model_name": "gpt-5.6",
+            "litellm_params": {"model": "openai/gpt-5.6", "input_cost_per_token": 3e-06},
+            "model_info": {"id": "dep-typed", "db_model": True},
+        },
+        {
+            "model_name": "gpt-5.6",
+            "litellm_params": {"model": "openai/gpt-5.6"},
+            "model_info": {"id": "dep-synced", "db_model": True},
+        },
+    ]
+    router: Final = MagicMock()
+    router.model_list = model_list
+    router.get_discovered_model_info = MagicMock(return_value={})
+    monkeypatch.setattr(proxy_server, "llm_router", router)
+    monkeypatch.setattr(proxy_server, "llm_model_list", model_list)
+    monkeypatch.setattr(proxy_server, "prisma_client", MagicMock())
+    monkeypatch.setattr(proxy_server, "user_model", None)
+    monkeypatch.setattr(proxy_server.proxy_config, "get_config", AsyncMock(return_value={}))
+    monkeypatch.setattr(
+        proxy_server,
+        "_apply_search_filter_to_models",
+        AsyncMock(side_effect=lambda all_models, **kw: (all_models, len(all_models))),
+    )
+    import litellm.proxy.agent_endpoints.model_list_helpers as mlh
+
+    monkeypatch.setattr(mlh, "append_agents_to_model_info", AsyncMock(side_effect=lambda models, **kw: models))
+
+    with auth_as():
+        response = client.get("/v2/model/info")
+
+    assert response.status_code == 200, response.text
+    by_id: Final = {m["model_info"]["id"]: m["model_info"] for m in response.json()["data"]}
+    assert by_id["dep-typed"]["pricing_overrides"] == ["input_cost_per_token"]
+    assert by_id["dep-typed"]["input_cost_per_token"] == 3e-06
+    assert by_id["dep-synced"]["pricing_overrides"] == []
+    assert by_id["dep-synced"]["input_cost_per_token"] == litellm.model_cost["gpt-5.6"]["input_cost_per_token"]
+
+
 def test_v1_model_info_star_wildcard_filter_keeps_provider_expansion(monkeypatch):
     from litellm.proxy._types import SpecialModelNames, UserAPIKeyAuth
     from litellm.proxy.auth import model_checks