fix(model-edit): don't clear the required model field via a null patch

Review follow-up: the generalized null-clear also removed litellm_params.model
when a caller sent it as an explicit null, leaving a modelless row that breaks
deployment reconstruction. Exempt `model` from the clear so a null there is
ignored while other nulled params are still cleared. Trim a redundant UI comment
the helper already documents.

litellm/proxy/management_endpoints/model_management_endpoints.py

@@ -386,14 +386,16 @@ def update_db_model(db_model: Deployment, updated_patch: updateDeployment) -> Pr
     #
     # A litellm_params field the caller explicitly sets to null is removed from the
     # stored params, so the UI can drop a param (e.g. reasoning_effort) instead of
-    # only overwriting it. The cross-blob mirror stays scoped to SPECIAL_MODEL_INFO_PARAMS
-    # (input/output cost per token/character and cache read/write costs), which
-    # Deployment.__init__ mirrors between litellm_params and model_info. The model_info
-    # loop below stays restricted to SPECIAL params so this path cannot null out
-    # privileged model_info fields like team_id or access groups.
+    # only overwriting it. `model` is exempt: it is required to reconstruct the
+    # deployment, so a null there is ignored rather than persisting a modelless row.
+    # The cross-blob mirror stays scoped to SPECIAL_MODEL_INFO_PARAMS (input/output
+    # cost per token/character and cache read/write costs), which Deployment.__init__
+    # mirrors between litellm_params and model_info. The model_info loop below stays
+    # restricted to SPECIAL params so this path cannot null out privileged model_info
+    # fields like team_id or access groups.
     if updated_patch.litellm_params:
         for field in updated_patch.litellm_params.model_fields_set:
-            if getattr(updated_patch.litellm_params, field) is None:
+            if field != "model" and getattr(updated_patch.litellm_params, field) is None:
                 merged_litellm_params.pop(field, None)
                 if field in SPECIAL_MODEL_INFO_PARAMS:
                     merged_model_info.pop(field, None)

tests/test_litellm/proxy/management_endpoints/test_model_management_endpoints.py

@@ -3296,6 +3296,25 @@ class TestUpdateDBModelClearParam:
         info = json.loads(result["model_info"])
         assert info.get("team_id") == "team-keep-me"
 
+    def test_null_model_field_is_not_cleared(self):
+        """`model` is required to rebuild the deployment, so an explicit null for it
+        must be ignored, not persisted as a modelless (invalid) row. Other nulled
+        params in the same patch are still cleared."""
+        from litellm.proxy.management_endpoints.model_management_endpoints import (
+            update_db_model,
+        )
+        from litellm.types.router import updateLiteLLMParams
+
+        result = update_db_model(
+            db_model=self._db_model(),
+            updated_patch=updateDeployment(
+                litellm_params=updateLiteLLMParams(model=None, reasoning_effort=None)
+            ),
+        )
+        params = json.loads(result["litellm_params"])
+        assert params.get("model") == "gpt-5.6-sol"
+        assert "reasoning_effort" not in params
+
 
 class TestGetModelInfoWithIdBlocked:
     """`ProxyConfig.get_model_info_with_id` must propagate the DB-level `blocked`

ui/litellm-dashboard/src/components/model_info_view.tsx

@@ -504,9 +504,8 @@ export default function ModelInfoView({
         return;
       }
 
-      // A param the user deleted from the editor is absent from updatedLitellmParams;
-      // send it as an explicit null so the backend clears it, since the PATCH merge is
-      // additive and would otherwise keep the stored value.
+      // The additive PATCH merge keeps omitted keys, so a deleted param needs an
+      // explicit null to actually clear on the backend.
       for (const removedKey of computeRemovedLitellmParamKeys(
         localModelData.litellm_params ?? {},
         updatedLitellmParams,