Share your thoughts, 1 month free Claude Pro on usSee more

Long-context understanding on LongBench (Specific Subtask Scores)

56.02MQA-E Score

CSAttention

Updated 5d ago

Evaluation Results

Method	Links
CSAttention 2026.03		56.02	62.01	30.46	26.38	31.11	71.5	44.16	91.95	99.5	55.94	33.6	63.33	45	17.63	52.04
Full 2026.03		55.54	62.87	29.91	27.16	30.89	72.5	43.75	91.65	100	56.16	35.26	64.89	46	17.16	52.41
Full 2026.03		53.67	63.37	26.05	24.88	36.18	71.5	44.3	88.54	100	59.4	33.35	69.13	47.5	14.31	52.3
CSAttention 2026.03		53.21	63.74	26.18	24.6	37.05	72	45.1	88.62	100	59.18	32.63	68.89	46	14.29	52.25
PQCache 2026.03		52.96	57.57	30.14	16.67	28.69	71	40.01	91.82	99	55.22	34.01	60.52	43	16.38	49.79
MagicPig 2026.03		52.11	57.32	26.31	18.94	28.08	58	44.22	87.9	98.5	51.88	24.32	55.67	40	9.12	46.6
PQCache 2026.03		51.98	60.35	26.98	21.9	36.9	72	42.8	84.1	100	58.73	33.09	61.01	44	14.39	50.59
Full 2026.03		50.21	53.19	27.74	26.57	26.5	70	46.3	89.04	97	51.08	34.22	64.32	47	15.68	49.92
H2O 2026.03		50.1	57.9	26.01	23.99	34.2	61	44.5	85	98.5	54.12	29.99	63.01	31	13.21	48.04
CSAttention 2026.03		49.92	52.94	25.56	27.06	26.1	70.5	45.91	90.59	97	49.34	32.88	63.98	46	16.44	49.92
MagicPig 2026.03		48.78	53.16	25.86	14.5	19.2	70	42	65.05	96	38.5	23.39	61.11	38	7.83	43.1
MagicPig 2026.03		45.87	38.91	26.01	23.34	21.42	71	45.02	90.15	95	34.98	31.29	55.04	29	14.08	44.37
PQCache 2026.03		45.57	39.59	22.57	26.04	22.3	71	42.18	88.62	89	35.22	29.68	64.01	46	15.01	45.49
SparQ 2026.03		45.32	50.07	25.88	21.03	31.9	59	43.9	80.2	91	47.93	27.62	48.32	25.5	13.48	43.65
H2O 2026.03		40.17	40.01	29.21	23.94	28.08	62	41.1	90.32	97	53.36	28.32	57.96	23.5	16.5	45.11
SparQ 2026.03		39.56	34.32	26.96	21.78	28.48	47	42.11	89.26	87	51.83	25.21	55.42	21	15.18	41.79
H2O 2026.03		37.26	30.43	21.07	25.33	17.01	63	41.98	84.77	52	31.56	22.92	59.91	31	6.04	37.45
SparQ 2026.03		31.51	31.77	19.62	21.86	15.63	61	41.68	84.1	42	29.69	25.67	53.01	34	5.81	35.53