fix(utils): honor string drop_params values from config and DB deployments

litellm/litellm_core_utils/core_helpers.py

@@ -36,6 +36,18 @@ def safe_divide_seconds(seconds: float, denominator: float, default: Optional[fl
     return float(seconds / denominator)
 
 
+def normalize_drop_params(value: object) -> bool | None:
+    if isinstance(value, bool):
+        return value
+    if isinstance(value, str):
+        lowered = value.strip().lower()
+        if lowered == "true":
+            return True
+        if lowered == "false":
+            return False
+    return None
+
+
 def safe_divide(
     numerator: Union[int, float],
     denominator: Union[int, float],

litellm/types/router.py

@@ -23,6 +23,7 @@ from pydantic import BaseModel, ConfigDict, Field, field_validator, model_valida
 from typing_extensions import Protocol, Required, TypedDict, runtime_checkable
 
 from litellm._uuid import uuid
+from litellm.litellm_core_utils.core_helpers import normalize_drop_params
 
 from .completion import CompletionRequest
 from .embedding import EmbeddingRequest
@@ -233,6 +234,7 @@ class GenericLiteLLMParams(CredentialLiteLLMParams, CustomPricingLiteLLMParams):
         None  # timeout when making stream=True calls, if str, pass in as os.environ/
     )
     max_retries: Optional[int] = None
+    drop_params: Optional[bool] = None
     organization: Optional[str] = None  # for openai orgs
     configurable_clientside_auth_params: CONFIGURABLE_CLIENTSIDE_AUTH_PARAMS = None
     litellm_credential_name: Optional[str] = None
@@ -311,6 +313,11 @@ class GenericLiteLLMParams(CredentialLiteLLMParams, CustomPricingLiteLLMParams):
             return filtered
         return data
 
+    @field_validator("drop_params", mode="before")
+    @classmethod
+    def coerce_drop_params(cls, value: object) -> Optional[bool]:
+        return normalize_drop_params(value)
+
     def __contains__(self, key):
         # Define custom behavior for the 'in' operator
         return hasattr(self, key)

litellm/utils.py

@@ -60,6 +60,7 @@ from litellm._lazy_imports import (
     _get_token_counter_new,
 )
 from litellm._uuid import uuid
+from litellm.litellm_core_utils.core_helpers import normalize_drop_params
 from litellm.litellm_core_utils.fallback_generalizations import (
     match_capability_generalizations,
 )
@@ -2852,7 +2853,7 @@ def get_optional_params_transcription(
 
     passed_params.pop("OPENAI_TRANSCRIPTION_PARAMS")
     custom_llm_provider = passed_params.pop("custom_llm_provider")
-    drop_params = passed_params.pop("drop_params")
+    drop_params = normalize_drop_params(passed_params.pop("drop_params"))
     special_params = passed_params.pop("kwargs")
     for k, v in special_params.items():
         passed_params[k] = v
@@ -2960,7 +2961,7 @@ def get_optional_params_image_gen(
     model = passed_params.pop("model", None)
     custom_llm_provider = passed_params.pop("custom_llm_provider")
     provider_config = passed_params.pop("provider_config", None)
-    drop_params = passed_params.pop("drop_params", None)
+    drop_params = normalize_drop_params(passed_params.pop("drop_params", None))
     additional_drop_params = passed_params.pop("additional_drop_params", None)
     special_params = passed_params.pop("kwargs")
     for k, v in special_params.items():
@@ -3084,7 +3085,7 @@ def get_optional_params_embeddings(
     custom_llm_provider = passed_params.pop("custom_llm_provider", None)
     special_params = passed_params.pop("kwargs")
 
-    drop_params = passed_params.pop("drop_params", None)
+    drop_params = normalize_drop_params(passed_params.pop("drop_params", None))
     additional_drop_params = passed_params.pop("additional_drop_params", None)
     allowed_openai_params = passed_params.pop("allowed_openai_params", None) or []
     # Remove function objects from passed_params to avoid JSON serialization errors
@@ -3797,6 +3798,8 @@ def get_optional_params(
 ):
     passed_params = locals().copy()
     special_params = passed_params.pop("kwargs")
+    drop_params = normalize_drop_params(drop_params)
+    passed_params["drop_params"] = drop_params
     # Remove base_model from passed_params so it doesn't interfere with
     # non_default_params / _check_valid_arg — it's a routing hint, not an
     # OpenAI param.

tests/test_litellm/litellm_core_utils/test_core_helpers.py

@@ -5,6 +5,7 @@ import pytest
 from litellm.litellm_core_utils.core_helpers import (
     _FINISH_REASON_MAP,
     map_finish_reason,
+    normalize_drop_params,
     reconstruct_model_name,
     redact_nested_match_and_regex_keys,
 )
@@ -201,3 +202,24 @@ class TestRedactNestedMatchAndRegexKeys:
     def test_passes_through_none_and_str(self):
         assert redact_nested_match_and_regex_keys(None) is None
         assert redact_nested_match_and_regex_keys("plain") == "plain"
+
+
+@pytest.mark.parametrize(
+    "value, expected",
+    [
+        (True, True),
+        (False, False),
+        ("true", True),
+        ("True", True),
+        (" TRUE ", True),
+        ("false", False),
+        ("False", False),
+        (None, None),
+        ("yes", None),
+        ("", None),
+        (1, None),
+        (0, None),
+    ],
+)
+def test_normalize_drop_params(value, expected):
+    assert normalize_drop_params(value) is expected

tests/test_litellm/test_router.py

@@ -5430,3 +5430,37 @@ class TestRouterRequestTimeoutPropagation:
             )
             == 60
         )
+
+
+@pytest.mark.asyncio
+async def test_router_deployment_drop_params_string_true_is_honored(monkeypatch):
+    from litellm import Router
+
+    monkeypatch.setattr(litellm, "drop_params", False)
+    router = Router(
+        model_list=[
+            {
+                "model_name": "gpt-5-nano",
+                "litellm_params": {
+                    "model": "openai/gpt-5-nano",
+                    "api_key": "sk-fake",
+                    "temperature": 1,
+                    "reasoning_effort": "minimal",
+                    "drop_params": "true",
+                    "mock_response": "Hello, world!",
+                },
+            }
+        ],
+        num_retries=0,
+    )
+
+    deployment = router.get_deployment_by_model_group_name(model_group_name="gpt-5-nano")
+    assert deployment is not None
+    assert deployment.litellm_params.drop_params is True
+
+    response = await router.acompletion(
+        model="gpt-5-nano",
+        messages=[{"role": "user", "content": "hi"}],
+        temperature=0.1,
+    )
+    assert response.choices[0].message.content == "Hello, world!"

tests/test_litellm/test_utils.py

@@ -4814,3 +4814,31 @@ def test_is_prompt_caching_valid_prompt_explicit_min_token_count_overrides_model
         is_prompt_caching_valid_prompt(model="claude-opus-4-8", messages=PROMPT_CACHE_MESSAGES, min_token_count=8192)
         is False
     )
+
+
+class TestDropParamsStringCoercion:
+    @pytest.mark.parametrize("drop_params", ["true", "True", True])
+    def test_truthy_drop_params_drops_unsupported_temperature(self, drop_params, monkeypatch):
+        from litellm.utils import get_optional_params
+
+        monkeypatch.setattr(litellm, "drop_params", False)
+        result = get_optional_params(
+            model="gpt-5-nano",
+            custom_llm_provider="openai",
+            temperature=0.1,
+            drop_params=drop_params,
+        )
+        assert "temperature" not in result
+
+    @pytest.mark.parametrize("drop_params", ["false", False, None])
+    def test_falsy_drop_params_still_raises(self, drop_params, monkeypatch):
+        from litellm.utils import get_optional_params
+
+        monkeypatch.setattr(litellm, "drop_params", False)
+        with pytest.raises(litellm.UnsupportedParamsError):
+            get_optional_params(
+                model="gpt-5-nano",
+                custom_llm_provider="openai",
+                temperature=0.1,
+                drop_params=drop_params,
+            )

ui/litellm-dashboard/src/lib/http/schema.d.ts

@@ -25667,6 +25667,8 @@ export interface components {
             default_api_key_rpm_limit?: number | null;
             /** Default Api Key Tpm Limit */
             default_api_key_tpm_limit?: number | null;
+            /** Drop Params */
+            drop_params?: boolean | null;
             /** Gcs Bucket Name */
             gcs_bucket_name?: string | null;
             /** Input Cost Per Audio Per Second */
@@ -33503,6 +33505,8 @@ export interface components {
             default_api_key_rpm_limit?: number | null;
             /** Default Api Key Tpm Limit */
             default_api_key_tpm_limit?: number | null;
+            /** Drop Params */
+            drop_params?: boolean | null;
             /** Gcs Bucket Name */
             gcs_bucket_name?: string | null;
             /** Input Cost Per Audio Per Second */