Long-context Language Understanding on LongBench (Specific Metric Subset including NQA and Delta Avg) benchmark leaderboard · SOTA2 Research

Long-context Language Understanding on LongBench (Specific Metric Subset including NQA and Delta Avg)

34.92NQAH2O

Evaluation Results

Method	Links
H2O2026.04		34.92	54.31	64.45	47.08	32.53	34.92	98	69.73	54.49	-0.15
Full Attention2026.04		34.81	54.24	64.45	47.08	32.7	34.93	98	70.88	54.64	—
SinkRouter2026.04		34.43	54.67	64.35	46.75	33.01	34.84	98	70.46	54.56	-0.08
FastGen2026.04		30.24	52.32	62.96	41.28	19.62	31.62	95.29	68.01	50.17	-4.47
SinkRouter2026.04		30.08	53.96	55.71	29.62	32.14	33.87	99.5	64.24	49.89	-0.69
Full Attention2026.04		30	55.14	56.15	30.08	33.85	35.2	99.5	64.73	50.58	—
H2O2026.04		29.69	54.99	56.1	30.55	34.11	35.15	99.5	64.81	50.61	0.03
FastGen2026.04		26.02	52.92	52.79	26.54	18.75	31.61	92.5	64.21	45.67	-4.91
StreamingLLM2026.04		25.35	46.42	55.23	26.15	16.39	33.41	60.5	68.95	41.55	-13.09
StreamingLLM2026.04		20.7	38.05	40.18	15.9	14.93	31.29	36.5	65.03	32.82	-17.76
H2O2026.04		12.25	36.56	50.73	26.95	21.04	29.3	58	72.84	38.46	-0.81
Full Attention2026.04		11.92	37.6	51.66	27.55	21.66	30.62	59.5	73.62	39.27	—
SinkRouter2026.04		11.66	37.24	52.01	27.81	20.81	30.13	50.5	73.22	37.92	-1.35
FastGen2026.04		10.65	34.29	43.97	17.52	12.94	24.43	33.5	71.87	31.15	-8.12
StreamingLLM2026.04		4.29	31.17	43.07	17.93	13.75	22.61	19.5	62.83	26.89	-12.38