Long-context language understanding on LongBench 1 host v1 (test) benchmark leaderboard · SOTA2 Research

Long-context language understanding on LongBench 1 host v1 (test)

46.232WQA Scorekvtc16x

Evaluation Results

Method	Links
kvtc16x2025.11		46.23	42.62	27.12	23.1	35.91	57.3
Vanilla2025.11		45.83	42.62	27.42	23.4	36.11	58.1
kvtc16x2025.11		43.73	51.02	26.62	25.2	45.91	52.2
Vanilla2025.11		43.33	51.52	27.22	25.7	45.61	52.8
KIVI2025.11		42.13	50.12	31.73	25.7	45.81	58.8
TOVA2025.11		41.33	50.42	32.63	25.4	46.11	54.8
Vanilla2025.11		40.83	50.32	33.83	26.9	47.31	59.9
GEAR2025.11		40.73	49.62	33.83	27.2	47.21	59.7
FP82025.11		40.63	50.92	33.83	26.1	47.11	59.9
kvtc64x2025.11		40.63	50.22	34.13	25.2	45.61	57.6
kvtc16x2025.11		40.33	51.12	33.73	26.4	47.01	59.2
kvtc32x2025.11		40.03	49.52	33.93	25.9	46.61	58.5
xKV2025.11		39.53	48.72	31.53	24.4	45.71	55.7
H2O2025.11		38.53	40.82	32.43	25.4	46.81	42