fix: bill Opper streams from the gateway's usage.cost

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AcQ2QoVJWZaePBLwGDDdpC
This commit is contained in:
Jose Sabater 2026-09-14 12:14:52 +02:00
parent 30648f2d2e
commit 95bfde26a7
2 changed files with 8 additions and 1 deletions

View file

@ -54,7 +54,9 @@ FUNCTION_CALL_ATTRIBUTE: Final = "function_call"
_SYNC_ITER_EXHAUSTED: Final = object()
_GCHUNK_FIELDS: Final[frozenset] = frozenset(GChunk.__annotations__)
_USAGE_COST_HEADER_PROVIDERS: Final[frozenset[str]] = frozenset({LlmProviders.OPENROUTER.value})
_USAGE_COST_HEADER_PROVIDERS: Final[frozenset[str]] = frozenset(
{LlmProviders.OPENROUTER.value, LlmProviders.OPPER.value}
)
def _next_sync_or_exhausted(it: Any) -> object:

View file

@ -231,6 +231,11 @@ def test_opper_streaming_dispatch_carries_cost(respx_mock, monkeypatch):
complete_response = litellm.stream_chunk_builder(response.chunks, messages=[{"role": "user", "content": "Hello"}])
assert complete_response.usage.cost == 3.3e-05
from litellm.litellm_core_utils.streaming_handler import CustomStreamWrapper
CustomStreamWrapper._propagate_usage_cost_to_hidden_params(complete_response, "opper")
assert complete_response._hidden_params["additional_headers"]["llm_provider-x-litellm-response-cost"] == 3.3e-05
@pytest.mark.respx()
def test_opper_reasoning_effort_passes_through(respx_mock, monkeypatch):