From 2fa7dfc6ba99f9fd46a3a2716309b31f64c81ce5 Mon Sep 17 00:00:00 2001
From: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
Date: Wed, 28 Oct 2026 19:45:52 -0700
Subject: [PATCH] fix(kyma): let SDK custom_cost_per_token and
 custom_cost_per_second override Kyma's usage.cost

---
 litellm/cost_calculator.py                    |  5 ++-
 .../llms/openai_like/test_kyma_provider.py    | 33 +++++++++++++++++++
 2 files changed, 37 insertions(+), 1 deletion(-)

diff --git a/litellm/cost_calculator.py b/litellm/cost_calculator.py
index 7821a96df52..eaeca40d5c5 100644
--- a/litellm/cost_calculator.py
+++ b/litellm/cost_calculator.py
@@ -1649,6 +1649,9 @@ def completion_cost(
             )
 
         explicit_pricing: Final = custom_pricing is True or base_model is not None
+        caller_set_prices: Final = (
+            explicit_pricing or custom_cost_per_token is not None or custom_cost_per_second is not None
+        )
         custom_priced_request_model: Final = model if custom_pricing is True else None
         configured_cost_per_query: Final = (
             _configured_search_cost_per_query(litellm_logging_obj) if custom_pricing else None
@@ -2019,7 +2022,7 @@ def completion_cost(
                 )
                 model_for_discount = custom_priced_request_model or model
                 if (
-                    not explicit_pricing
+                    not caller_set_prices
                     and (reported_cost := billed_usage_cost(priced_response, custom_llm_provider)) is not None
                 ):
                     return _adjusted_provider_billed_cost(
diff --git a/tests/unit/llms/openai_like/test_kyma_provider.py b/tests/unit/llms/openai_like/test_kyma_provider.py
index a05ca10713f..2b3740ca676 100644
--- a/tests/unit/llms/openai_like/test_kyma_provider.py
+++ b/tests/unit/llms/openai_like/test_kyma_provider.py
@@ -292,6 +292,39 @@ def test_sdk_completion_cost_prices_kyma_at_the_cost_kyma_reports():
     assert litellm.completion_cost(completion_response=response) == pytest.approx(reported_cost)
 
 
+def _kyma_sdk_response(usage: dict[str, float]) -> litellm.ModelResponse:
+    with respx.mock() as upstream:
+        upstream.post(_CHAT_URL).respond(200, json=_kyma_chat_body(usage))
+        response: Final = litellm.completion(model="kyma/sonar", messages=[{"role": "user", "content": "ping"}], api_key="k")
+    assert isinstance(response, litellm.ModelResponse)
+    return response
+
+
+_KYMA_TOKEN_USAGE: Final = {"prompt_tokens": 4, "completion_tokens": 3, "total_tokens": 7}
+_KYMA_REPORTED_COST: Final = 0.00721
+
+
+def test_sdk_completion_cost_uses_the_callers_token_prices_over_the_cost_kyma_reports():
+    caller_prices: Final = {"input_cost_per_token": 0.5, "output_cost_per_token": 2.0}
+    with_reported_cost: Final = _kyma_sdk_response({**_KYMA_TOKEN_USAGE, "cost": _KYMA_REPORTED_COST})
+    without_reported_cost: Final = _kyma_sdk_response(_KYMA_TOKEN_USAGE)
+
+    assert litellm.completion_cost(
+        completion_response=with_reported_cost, custom_cost_per_token=caller_prices
+    ) == pytest.approx(litellm.completion_cost(completion_response=without_reported_cost, custom_cost_per_token=caller_prices))
+    assert litellm.completion_cost(
+        completion_response=with_reported_cost, custom_cost_per_token=caller_prices
+    ) != pytest.approx(_KYMA_REPORTED_COST)
+
+
+def test_sdk_completion_cost_uses_the_callers_per_second_price_over_the_cost_kyma_reports():
+    response: Final = _kyma_sdk_response({**_KYMA_TOKEN_USAGE, "cost": _KYMA_REPORTED_COST})
+
+    assert litellm.completion_cost(completion_response=response, custom_cost_per_second=50.0) == pytest.approx(
+        50.0 * response._response_ms / 1000
+    )
+
+
 def _with_kyma_discount_and_margin(monkeypatch: pytest.MonkeyPatch) -> float:
     monkeypatch.setattr(litellm, "cost_discount_config", {"kyma": 0.5})
     monkeypatch.setattr(litellm, "cost_margin_config", {"kyma": 0.1})