diff --git a/litellm/responses/litellm_completion_transformation/transformation.py b/litellm/responses/litellm_completion_transformation/transformation.py index b2d1a69e0d8..2b011abf0e2 100644 --- a/litellm/responses/litellm_completion_transformation/transformation.py +++ b/litellm/responses/litellm_completion_transformation/transformation.py @@ -2667,6 +2667,14 @@ class LiteLLMCompletionResponsesConfig: if hasattr(prompt_details, "audio_tokens") and prompt_details.audio_tokens is not None: input_details_dict["audio_tokens"] = prompt_details.audio_tokens + # The cost path reads the grounding counters off the input details, and a realtime + # session's usage is rebuilt from its own response.done, so dropping them here bills + # no per-query grounding fee at all. + for counter in ("web_search_requests", "google_maps_grounding_requests"): + counter_value = getattr(prompt_details, counter, None) + if counter_value is not None: + input_details_dict[counter] = counter_value + cache_write_tokens = getattr(prompt_details, "cache_write_tokens", None) or getattr( prompt_details, "cache_creation_tokens", None ) diff --git a/litellm/responses/utils.py b/litellm/responses/utils.py index 540d492beec..7e0acd14b9c 100644 --- a/litellm/responses/utils.py +++ b/litellm/responses/utils.py @@ -1137,6 +1137,12 @@ class ResponseAPILoggingUtils: text_tokens=getattr(response_api_usage.input_tokens_details, "text_tokens", None), image_tokens=getattr(response_api_usage.input_tokens_details, "image_tokens", None), cache_write_tokens=getattr(response_api_usage.input_tokens_details, "cache_write_tokens", None), + web_search_requests=getattr( + response_api_usage.input_tokens_details, "web_search_requests", None + ), + google_maps_grounding_requests=getattr( + response_api_usage.input_tokens_details, "google_maps_grounding_requests", None + ), ) completion_tokens_details: CompletionTokensDetailsWrapper | None = None output_tokens_details: Final[OutputTokensDetails | None] = getattr(