diff --git a/litellm/litellm_core_utils/token_counter.py b/litellm/litellm_core_utils/token_counter.py index 69505309ef0..781b0892849 100644 --- a/litellm/litellm_core_utils/token_counter.py +++ b/litellm/litellm_core_utils/token_counter.py @@ -758,8 +758,10 @@ def pdf_page_pricing(model: str | None) -> PdfPagePricing: def reads_pdf_pages(model: str) -> bool: - entry: Final = _cost_map_entry(model) or _cost_map_entry(model.partition("/")[2]) - return entry.get("supports_pdf_input") is True + segments: Final = model.split("/") + names: Final = ("/".join(segments[start:]) for start in range(len(segments))) + stated: Final = (_cost_map_entry(name).get("supports_pdf_input") for name in names) + return next((support for support in stated if support is not None), False) is True def token_counter( diff --git a/tests/unit/proxy/spend_tracking/test_input_tokens.py b/tests/unit/proxy/spend_tracking/test_input_tokens.py index 148001be32b..1a1cbf513ab 100644 --- a/tests/unit/proxy/spend_tracking/test_input_tokens.py +++ b/tests/unit/proxy/spend_tracking/test_input_tokens.py @@ -75,8 +75,18 @@ def _router(*deployments: tuple[str, str]) -> Router: @pytest.mark.parametrize( ("group", "deployment_model"), - [("gpt-4o", "anthropic/claude-haiku-5-5"), ("pdf-reports", "azure/gpt-5.6-sol")], - ids=["gpt-named-group-served-by-claude", "custom-group-served-by-azure-gpt"], + [ + ("gpt-4o", "anthropic/claude-haiku-5-5"), + ("pdf-reports", "azure/gpt-5.6-sol"), + ("pdf-reports", "azure/responses/gpt-5.6-sol"), + ("pdf-reports", "openrouter/openai/gpt-5.6-sol"), + ], + ids=[ + "gpt-named-group-served-by-claude", + "custom-group-served-by-azure-gpt", + "custom-group-served-by-azure-responses-gpt", + "custom-group-served-by-openrouter-gpt", + ], ) def test_a_model_group_counts_like_the_deployment_that_serves_it(group: str, deployment_model: str) -> None: request_body: Final = {"model": group, "messages": _one_page_pdf_message()}