Merge pull request #36215 from BerriAI/litellm_declare_model_info_pricing

refactor(types): declare mirrored pricing fields on ModelInfo

litellm/types/router.py

@@ -17,7 +17,12 @@ from .completion import CompletionRequest
 from .embedding import EmbeddingRequest
 from .llms.openai import OpenAIFileObject
 from .search import SearchProvider
-from .utils import CustomPricingLiteLLMParams, ModelResponse, StandardLoggingRoutingDecision
+from .utils import (
+    CustomPricingLiteLLMParams,
+    MirroredPricingParams,
+    ModelResponse,
+    StandardLoggingRoutingDecision,
+)
 
 
 class ConfigurableClientsideParamsCustomAuth(TypedDict):
@@ -122,7 +127,7 @@ class UpdateRouterConfig(BaseModel):
     model_config = ConfigDict(protected_namespaces=())
 
 
-class ModelInfo(BaseModel):
+class ModelInfo(MirroredPricingParams):
     id: str | None  # Allow id to be optional on input, but it will always be present as a str in the model instance
     db_model: bool = False  # used for proxy - to separate models which are stored in the db vs. config.
     updated_at: datetime.datetime | None = None
@@ -424,14 +429,7 @@ class DeploymentTypedDict(TypedDict, total=False):
     model_info: dict
 
 
-SPECIAL_MODEL_INFO_PARAMS = [
-    "input_cost_per_token",
-    "output_cost_per_token",
-    "input_cost_per_character",
-    "output_cost_per_character",
-    "cache_read_input_token_cost",
-    "cache_creation_input_token_cost",
-]
+SPECIAL_MODEL_INFO_PARAMS = tuple(MirroredPricingParams.model_fields)
 
 
 class Deployment(BaseModel):

litellm/types/utils.py

@@ -3245,10 +3245,23 @@ class StandardCallbackDynamicParams(TypedDict, total=False):
     litellm_disabled_callbacks: list[str] | None
 
 
-class CustomPricingLiteLLMParams(BaseModel):
-    ## CUSTOM PRICING ##
+class MirroredPricingParams(BaseModel):
+    """Pricing overrides that ``Deployment.__init__`` mirrors from ``litellm_params``
+    onto ``model_info``, so both blobs hold the same rate.
+
+    Declared once and inherited by both sides of that mirror, so the two can't drift.
+    """
+
     input_cost_per_token: float | None = None
     output_cost_per_token: float | None = None
+    input_cost_per_character: float | None = None
+    output_cost_per_character: float | None = None
+    cache_read_input_token_cost: float | None = None
+    cache_creation_input_token_cost: float | None = None
+
+
+class CustomPricingLiteLLMParams(MirroredPricingParams):
+    ## CUSTOM PRICING ##
     input_cost_per_second: float | None = None
     output_cost_per_second: float | None = None
     output_cost_per_second_1080p: float | None = None
@@ -3259,7 +3272,6 @@ class CustomPricingLiteLLMParams(BaseModel):
     # This allows any model_info parameter to be set in litellm_params
     input_cost_per_token_flex: float | None = None
     input_cost_per_token_priority: float | None = None
-    cache_creation_input_token_cost: float | None = None
     cache_creation_input_token_cost_above_1hr: float | None = None
     cache_creation_input_token_cost_above_200k_tokens: float | None = None
     cache_creation_input_token_cost_above_272k_tokens: float | None = None
@@ -3268,7 +3280,6 @@ class CustomPricingLiteLLMParams(BaseModel):
     cache_creation_input_token_cost_flex: float | None = None
     cache_creation_input_token_cost_priority: float | None = None
     cache_creation_input_audio_token_cost: float | None = None
-    cache_read_input_token_cost: float | None = None
     cache_read_input_token_cost_flex: float | None = None
     cache_read_input_token_cost_priority: float | None = None
     cache_read_input_token_cost_above_200k_tokens: float | None = None
@@ -3276,7 +3287,6 @@ class CustomPricingLiteLLMParams(BaseModel):
     cache_read_input_token_cost_above_272k_tokens_priority: float | None = None
     cache_read_input_token_cost_above_272k_tokens_flex: float | None = None
     cache_read_input_audio_token_cost: float | None = None
-    input_cost_per_character: float | None = None
     input_cost_per_character_above_128k_tokens: float | None = None
     input_cost_per_audio_token: float | None = None
     input_cost_per_token_cache_hit: float | None = None
@@ -3298,7 +3308,6 @@ class CustomPricingLiteLLMParams(BaseModel):
     output_cost_per_token_batches: float | None = None
     output_cost_per_token_flex: float | None = None
     output_cost_per_token_priority: float | None = None
-    output_cost_per_character: float | None = None
     output_cost_per_audio_token: float | None = None
     output_cost_per_token_above_128k_tokens: float | None = None
     output_cost_per_token_above_200k_tokens: float | None = None

tests/test_litellm/types/test_router.py

@@ -0,0 +1,73 @@
+import pytest
+
+from litellm.types.router import (
+    SPECIAL_MODEL_INFO_PARAMS,
+    Deployment,
+    LiteLLM_Params,
+    ModelInfo,
+)
+from litellm.types.utils import CustomPricingLiteLLMParams, MirroredPricingParams
+
+
+def test_model_info_declares_mirrored_pricing_fields():
+    """The pricing keys Deployment mirrors onto model_info must be declared fields, not
+    extras that only survive because ModelInfo sets extra="allow"."""
+    for field in SPECIAL_MODEL_INFO_PARAMS:
+        assert field in ModelInfo.model_fields
+
+    info = ModelInfo(id="x", input_cost_per_token=1e-06)
+    assert info.__pydantic_extra__ == {}
+    assert info.input_cost_per_token == 1e-06
+
+
+def test_special_model_info_params_cannot_drift_from_the_mirror():
+    assert SPECIAL_MODEL_INFO_PARAMS == tuple(MirroredPricingParams.model_fields)
+    assert set(SPECIAL_MODEL_INFO_PARAMS) <= set(CustomPricingLiteLLMParams.model_fields)
+    assert set(SPECIAL_MODEL_INFO_PARAMS) <= set(LiteLLM_Params.model_fields)
+
+
+def test_custom_pricing_params_keeps_every_field_it_had():
+    """The mirrored fields moved to a base class; none of them may go missing from
+    CustomPricingLiteLLMParams, whose model_fields drive custom-pricing detection."""
+    for field in (
+        "input_cost_per_token",
+        "output_cost_per_token",
+        "input_cost_per_character",
+        "output_cost_per_character",
+        "cache_read_input_token_cost",
+        "cache_creation_input_token_cost",
+        "input_cost_per_second",
+        "cache_read_input_token_cost_flex",
+        "input_cost_per_character_above_128k_tokens",
+        "output_cost_per_audio_token",
+    ):
+        assert field in CustomPricingLiteLLMParams.model_fields
+
+
+@pytest.mark.parametrize("field", SPECIAL_MODEL_INFO_PARAMS)
+def test_deployment_mirrors_pricing_from_litellm_params_onto_model_info(field):
+    deployment = Deployment(
+        model_name="my-model",
+        litellm_params=LiteLLM_Params(model="gpt-4o", **{field: 3e-06}),
+    )
+    assert getattr(deployment.model_info, field) == 3e-06
+    assert deployment.model_info.model_dump(exclude_none=True)[field] == 3e-06
+
+
+def test_unset_pricing_is_still_absent_from_dumps():
+    """/model/info responses and DB writes dump model_info with exclude_none=True, so
+    declaring the pricing fields must not start emitting ~6 null keys per deployment."""
+    dumped = ModelInfo(id="x").model_dump(exclude_none=True)
+    assert [field for field in SPECIAL_MODEL_INFO_PARAMS if field in dumped] == []
+
+
+def test_pricing_strings_are_coerced_to_float():
+    """Cost values arrive from the DB and the Admin UI as strings; they must land as
+    floats so cost calculation doesn't multiply a str."""
+    info = ModelInfo(id="x", output_cost_per_token="0.000002")
+    assert info.output_cost_per_token == 2e-06
+
+
+def test_invalid_pricing_is_rejected():
+    with pytest.raises(ValueError):
+        ModelInfo(id="x", input_cost_per_token="free")

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

@@ -35294,6 +35294,10 @@ export interface components {
             base_model?: string | null;
             /** Blocked */
             blocked?: boolean | null;
+            /** Cache Creation Input Token Cost */
+            cache_creation_input_token_cost?: number | null;
+            /** Cache Read Input Token Cost */
+            cache_read_input_token_cost?: number | null;
             /** Created At */
             created_at?: string | null;
             /** Created By */
@@ -35305,6 +35309,14 @@ export interface components {
             db_model: boolean;
             /** Id */
             id: string | null;
+            /** Input Cost Per Character */
+            input_cost_per_character?: number | null;
+            /** Input Cost Per Token */
+            input_cost_per_token?: number | null;
+            /** Output Cost Per Character */
+            output_cost_per_character?: number | null;
+            /** Output Cost Per Token */
+            output_cost_per_token?: number | null;
             /** Team Id */
             team_id?: string | null;
             /** Team Public Model Name */