Long-context language understanding on LongBench (7-Metric Subset: NQ, HQ, QM, MN, GR, TQ, Avg) benchmark leaderboard · SOTA2 Research

Long-context language understanding on LongBench (7-Metric Subset: NQ, HQ, QM, MN, GR, TQ, Avg)

29.52NQ ScoreDASH-KV

Evaluation Results

Method	Links
DASH-KV2026.04		29.52	61.83	45.79	14.44	18.06	87.88	42.92
Full Attn2026.04		28.2	61.98	45.52	14.19	16.78	86.91	42.26
Full Attn2026.04		28.11	57.43	45.29	15.2	19.97	90.22	42.7
DASH-KV2026.04		27.61	58.01	45.35	14.9	19.25	89.47	42.43
Full Attn2026.04		25.13	44.04	46.13	15.42	18.06	83.47	38.71
DASH-KV2026.04		24.65	44.5	45.34	15.4	19.13	83.33	38.73
H2O2026.04		24.4	18	27.6	20.99	18.96	89.7	33.28
SnapKV2026.04		24.1	20	34.6	20.9	18.97	90	34.76
SnapKV2026.04		23.86	15.6	38.61	23.07	24.56	89.31	35.84
H2O2026.04		23.1	16	21.3	23.4	22.3	90.2	32.72
H2O2026.04		22.88	13.3	34.28	22.72	23.69	88.75	34.27
SnapKV2026.04		21.3	16.6	30.8	26.3	22.2	90.2	34.57
StreamLLM2026.04		20.47	14.31	26.97	24.88	25.7	76.56	31.48
StreamLLM2026.04		18.3	16.4	23.4	16.04	19.01	74.1	27.87
StreamLLM2026.04		13.6	11.7	20.1	24.7	21.5	71.1	27.12