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

Long-context language understanding on LongBench 2024 (test)

47.23Multi-doc QA

Block-Dist - Full

Updated 15d ago

Evaluation Results

Method	Links
Block-Dist - Full 2026.05		47.23	47.75	22.87	63.45	70.05	17.71	44.84
Block-Dist - Block 2026.05		46.33	46.54	22.89	63.76	69.23	6.43	42.53
Original - Full 2026.05		46.25	48.86	23.94	62.44	69.17	5.66	42.72
Block-Dist - Full 2026.05		42.75	47.11	25.75	38.97	67.33	11.33	38.87
Block-Dist - Full 2026.05		42	47.42	25.88	25.67	66.5	20.11	37.93
Block-Dist - Block 2026.05		41.88	46.98	25.35	24.96	66.87	11.89	36.32
Block-Dist - Block 2026.05		41.77	46.38	25.17	36.76	67	9.95	37.83
Original - Full 2026.05		41.75	47.05	25.86	24.16	65.91	15.77	36.75
Original - Full 2026.05		40.52	47.51	24.01	35.59	66.99	1.66	36.05
Block-Dist - Full 2026.05		38.74	47.14	22.33	59.22	67.66	12.7	41.29
Block-Dist - Block 2026.05		37.55	46.17	24.14	59	66.32	11.98	40.86
Original - Full 2026.05		36.88	46.05	22.08	61.63	66.27	4.89	39.63
Original - Superposition 2026.05		35.88	32.85	22.77	48.54	45.23	10.41	32.61
Original - Prompt Cache 2026.05		33.28	39.87	22.41	19.45	50.76	14.83	30.1
Original - Prompt Cache 2026.05		32.48	35.8	24.58	52.83	49.42	8.64	33.96
Original - Superposition 2026.05		31.95	37.81	25.75	21.45	46.15	13.87	29.5
Original - Superposition 2026.05		29.75	35.86	28.96	22.78	49.65	8.51	29.25
Original - Prompt Cache 2026.05		28.89	38.86	23.77	20.76	45.27	5.67	27.2
Original - Prompt Cache 2026.05		28.36	31.89	17.89	49.23	48.49	4.83	30.12
Original - Superposition 2026.05		25.39	34.85	18.38	41.52	45.72	2.32	28.02