fix(ocr): bill request-level OCR pricing and fall back to the map without credits

litellm/cost_calculator.py

@@ -1948,12 +1948,13 @@ def ocr_cost(
     if response.usage_info is None:
         raise ValueError("OCR response usage_info is None")
 
+    credits: Final = getattr(response.usage_info, "credits", None)
     has_custom_ocr_pricing: Final = model_info is not None and (
-        model_info.get("ocr_cost_per_page") is not None or model_info.get("ocr_cost_per_credit") is not None
+        model_info.get("ocr_cost_per_page") is not None
+        or (credits is not None and model_info.get("ocr_cost_per_credit") is not None)
     )
     pricing: Final = model_info if has_custom_ocr_pricing else _cost_map_model_info(model, custom_llm_provider)
 
-    credits: Final = getattr(response.usage_info, "credits", None)
     cost_per_credit: Final = pricing.get("ocr_cost_per_credit") if pricing is not None else None
     if credits is not None and cost_per_credit is not None:
         return cost_per_credit * credits, 0.0

litellm/ocr/main.py

@@ -32,6 +32,7 @@ from litellm.rust_bridge import ocr as rust_ocr_bridge
 from litellm.rust_bridge.bindings import native_exception_types
 from litellm.rust_bridge.configuration import rust_enabled
 from litellm.types.router import GenericLiteLLMParams
+from litellm.types.utils import CustomPricingLiteLLMParams
 from litellm.utils import ProviderConfigManager, client
 
 ####### ENVIRONMENT VARIABLES ###################
@@ -171,6 +172,7 @@ def _prepare_ocr_request(
         litellm_params={
             "litellm_call_id": litellm_call_id,
             "api_base": api_base,
+            **litellm_params.model_dump(include=frozenset(CustomPricingLiteLLMParams.model_fields), exclude_none=True),
         },
         custom_llm_provider=custom_llm_provider,
     )

tests/test_litellm/ocr/test_main.py

@@ -0,0 +1,49 @@
+from typing import Final
+
+from litellm.litellm_core_utils.litellm_logging import Logging, use_custom_pricing_for_model
+from litellm.ocr.main import _prepare_ocr_request
+
+OCR_MODEL: Final = "mistral/mistral-ocr-4-1"
+DOCUMENT: Final = {"type": "document_url", "document_url": "https://example.com/doc.pdf"}
+
+
+def _logging_obj() -> Logging:
+    return Logging(
+        model=OCR_MODEL,
+        messages=[],
+        stream=False,
+        call_type="ocr",
+        start_time=None,
+        litellm_call_id="test-ocr-request-pricing",
+        function_id="1234",
+    )
+
+
+def _prepare(kwargs: dict[str, object]) -> Logging:
+    logging_obj: Final = _logging_obj()
+    _prepare_ocr_request(
+        model=OCR_MODEL,
+        document=dict(DOCUMENT),
+        api_key="test-key",
+        api_base=None,
+        timeout=None,
+        custom_llm_provider=None,
+        extra_headers=None,
+        kwargs={"litellm_logging_obj": logging_obj, **kwargs},
+    )
+    return logging_obj
+
+
+def test_prepare_ocr_request_forwards_custom_pricing_to_logging_params() -> None:
+    logging_obj: Final = _prepare({"ocr_cost_per_page": 0.05, "ocr_cost_per_credit": 0.5})
+
+    assert logging_obj.litellm_params["ocr_cost_per_page"] == 0.05
+    assert logging_obj.litellm_params["ocr_cost_per_credit"] == 0.5
+    assert use_custom_pricing_for_model(logging_obj.litellm_params) is True
+
+
+def test_prepare_ocr_request_without_custom_pricing_leaves_logging_params_unpriced() -> None:
+    logging_obj: Final = _prepare({})
+
+    assert "ocr_cost_per_page" not in logging_obj.litellm_params
+    assert use_custom_pricing_for_model(logging_obj.litellm_params) is False

tests/test_litellm/test_cost_calculator.py

@@ -4746,6 +4746,22 @@ def test_ocr_cost_falls_through_to_cost_map_when_deployment_has_no_ocr_pricing()
     assert cost == pytest.approx(map_price * 2)
 
 
+@pytest.mark.usefixtures("_local_model_cost_map")
+def test_ocr_cost_ignores_deployment_credit_pricing_when_response_reports_no_credits():
+    from litellm.cost_calculator import ocr_cost
+
+    map_price: Final = litellm.get_model_info(MAPPED_OCR_MODEL)["ocr_cost_per_page"]
+    assert map_price is not None
+
+    cost, _ = ocr_cost(
+        model=MAPPED_OCR_MODEL,
+        custom_llm_provider="mistral",
+        response=_ocr_response(MAPPED_OCR_MODEL, pages_processed=2),
+        model_info={"ocr_cost_per_credit": 0.5},
+    )
+    assert cost == pytest.approx(map_price * 2)
+
+
 @pytest.mark.parametrize("metadata_key", ["metadata", "litellm_metadata"])
 def test_completion_cost_ocr_reads_deployment_pricing_from_logging_metadata(metadata_key: str):
     logging_obj = _ocr_logging_obj({metadata_key: {"model_info": {"ocr_cost_per_page": 0.004}}})