@@ -850,6 +850,123 @@ def test_convert_responses_usage_derives_total_when_absent() -> None:
850850 assert result ["total_tokens" ] == 12
851851
852852
853+ def test_convert_responses_usage_maps_cache_token_details () -> None :
854+ details = _make_response_obj (
855+ cache_read_input_tokens = 800 , cache_creation_input_tokens = 150
856+ )
857+ usage = _make_response_obj (
858+ input_tokens = 1000 ,
859+ output_tokens = 50 ,
860+ total_tokens = 1050 ,
861+ input_tokens_details = details ,
862+ )
863+ result = _convert_responses_usage (usage )
864+ assert result is not None
865+ assert result ["input_token_details" ] == {"cache_read" : 800 , "cache_creation" : 150 }
866+
867+
868+ def test_convert_responses_usage_maps_cache_token_details_from_mapping () -> None :
869+ """`_get_attr` accepts dict payloads, so a mapping must map identically."""
870+ usage = {
871+ "input_tokens" : 1000 ,
872+ "output_tokens" : 50 ,
873+ "total_tokens" : 1050 ,
874+ "input_tokens_details" : {
875+ "cache_read_input_tokens" : 800 ,
876+ "cache_creation_input_tokens" : 150 ,
877+ },
878+ }
879+ result = _convert_responses_usage (usage )
880+ assert result is not None
881+ assert result ["input_token_details" ] == {"cache_read" : 800 , "cache_creation" : 150 }
882+
883+
884+ def test_convert_responses_usage_maps_only_reported_cache_fields () -> None :
885+ details = _make_response_obj (
886+ cache_read_input_tokens = 800 , cache_creation_input_tokens = None
887+ )
888+ usage = _make_response_obj (
889+ input_tokens = 1000 ,
890+ output_tokens = 50 ,
891+ total_tokens = 1050 ,
892+ input_tokens_details = details ,
893+ )
894+ result = _convert_responses_usage (usage )
895+ assert result is not None
896+ assert result ["input_token_details" ] == {"cache_read" : 800 }
897+
898+
899+ def test_convert_responses_usage_omits_cache_details_when_not_reported () -> None :
900+ usage = _make_response_obj (input_tokens = 10 , output_tokens = 2 , total_tokens = 12 )
901+ result = _convert_responses_usage (usage )
902+ assert result is not None
903+ assert result ["input_token_details" ] == {}
904+
905+
906+ def test_convert_responses_to_chat_result_includes_cache_token_details () -> None :
907+ details = _make_response_obj (
908+ cache_read_input_tokens = 800 , cache_creation_input_tokens = 150
909+ )
910+ usage = _make_response_obj (
911+ input_tokens = 1000 ,
912+ output_tokens = 50 ,
913+ total_tokens = 1050 ,
914+ input_tokens_details = details ,
915+ )
916+ response = _make_response_obj (
917+ id = "resp_cache" ,
918+ model = "sonar-pro" ,
919+ status = "completed" ,
920+ object = "response" ,
921+ output_text = "hi" ,
922+ output = [],
923+ usage = usage ,
924+ citations = None ,
925+ images = None ,
926+ related_questions = None ,
927+ search_results = None ,
928+ )
929+
930+ result = _convert_responses_to_chat_result (response )
931+ message = result .generations [0 ].message
932+
933+ assert isinstance (message , AIMessage )
934+ assert message .usage_metadata is not None
935+ assert message .usage_metadata ["input_token_details" ] == {
936+ "cache_read" : 800 ,
937+ "cache_creation" : 150 ,
938+ }
939+
940+
941+ def test_stream_event_conversion_for_completed_includes_cache_token_details () -> None :
942+ details = _make_response_obj (
943+ cache_read_input_tokens = 800 , cache_creation_input_tokens = 150
944+ )
945+ usage = _make_response_obj (
946+ input_tokens = 1000 ,
947+ output_tokens = 50 ,
948+ total_tokens = 1050 ,
949+ input_tokens_details = details ,
950+ )
951+ response = _make_response_obj (
952+ id = "resp_cache_stream" ,
953+ model = "sonar-pro" ,
954+ status = "completed" ,
955+ object = "response" ,
956+ usage = usage ,
957+ )
958+ chunk = _convert_responses_stream_event_to_chunk (
959+ _make_event ("response.completed" , response = response )
960+ )
961+ assert chunk is not None
962+ assert isinstance (chunk .message , AIMessageChunk )
963+ assert chunk .message .usage_metadata is not None
964+ assert chunk .message .usage_metadata ["input_token_details" ] == {
965+ "cache_read" : 800 ,
966+ "cache_creation" : 150 ,
967+ }
968+
969+
853970# ---------------------------------------------------------------------------
854971# Error and edge cases in conversion / streaming
855972# ---------------------------------------------------------------------------
0 commit comments