Long-context language understanding on LongBench (16 Subtask Metrics + Avg)

31.42NQAFull KV

Evaluation results

Method
Full KV2026.05	31.42	43.15	55.84	51.72	41.56	30.24	34.62	27.58	25.41	78.52	90.12	49.82	7.15	99.2	54.32	55.76	48.53
Meta-Soft2026.05	31.34	34.28	55.43	50.46	41.56	30.12	27.42	27.84	25.41	71.75	90.12	49.82	7.15	99.2	54.32	54.62	47.15
Full KV2026.05	30.62	46.45	57.34	58.21	50.12	32.42	35.15	25.54	27.91	73.42	92.68	43.12	8.42	100	62.14	52.58	49.76
Meta-Soft2026.05	30.4	30.24	54.16	49.1	41.56	30.24	25.26	27.58	25.32	69.14	90.12	45.98	7.15	99.2	54.32	54.11	45.77
AnDPro2026.05	30.39	32.97	51.93	50.66	41.15	30.85	27.72	28.09	24.45	69.02	90.12	49.82	7.15	99.2	54.32	53.59	46.82
Judge Q2026.05	30.27	33.38	52.55	51.54	41.5	30.61	27.84	28.57	24.77	65.68	90.12	49.82	7.15	99.2	54.32	53.69	46.09
AnDPro2026.05	29.9	30.23	50.58	50.18	40.73	30.24	25.78	27.58	24.45	66.06	90.12	48.89	7.15	99.2	54.22	53.84	45.54
LAQ2026.05	29.54	30.6	52.57	51.92	41.98	31.12	27.37	30.36	24.93	66.83	90.12	49.82	7.15	99.2	53.69	52.74	45.24
H2O2026.05	29.4	29.9	43.15	47.35	39.02	29.07	27.52	28.79	23.65	59.02	90.12	49.82	7.14	99.2	54.32	54.4	43.68
Meta-Soft2026.05	29.1	35.29	55.22	54.98	49.39	31.72	27.3	25.24	25.65	66.55	92.5	39.95	8.21	99.64	61.64	52.19	47.19
SnapKV2026.05	29.06	35.17	53.61	51.97	41.09	29	27.6	28.28	23.44	61.86	90.12	49.82	7.15	99.2	53.54	53.77	45.04
Judge Q2026.05	28.7	29.09	50.29	50.31	40.2	29.52	24.56	27.47	23.36	61.12	90.02	49.43	7.15	99.2	53.45	53.41	44.83
AnDPro2026.05	28.62	34.38	52.19	55.73	48.27	31.95	27.07	24.99	25.67	63.75	92.24	41.79	8.03	99.89	61.1	54.94	46.98
ZeroMerge2026.05	28.61	31.25	50.45	50.26	40.17	29.2	25.97	28.33	24.55	65.41	90.12	49.82	7.15	99.2	53.07	52.56	45.65
Meta-Soft2026.05	28.54	31.34	54.42	53.82	49.11	31.45	24.52	25.12	24.84	64.12	92.42	38.82	8.12	99.5	61.45	51.84	46.21
CaM2026.05	28.39	30.11	50.16	49.79	39.87	29.21	25.88	28.73	24.2	62.66	90.12	48.36	7.15	99.2	53.63	52.85	44.87
AnDPro2026.05	28.12	31.42	50.84	55.12	47.82	31.82	25.12	24.84	25.12	61.42	92.12	41.42	7.92	99.85	60.84	54.91	46.17
ZeroMerge2026.05	28.1	28.98	49.41	50.08	39.82	29.05	24.27	27.13	22.57	61.24	90.12	47.58	7.15	99.2	53.77	53.21	44.52
CaM2026.05	27.87	27.77	49.21	49.58	39.61	27.88	23.14	26.96	22.51	58.87	89.86	46.94	7.15	99.2	52.42	52.55	43.84
Judge Q2026.05	27.78	34.04	52.09	55.92	47.83	30.92	26.47	24.78	25.24	60.39	91.49	42.13	7.99	99.38	60.41	53.26	46.27
LAQ2026.05	27.4	26.36	48.27	49.38	38.44	27.73	22.27	27.14	21.24	58.18	89.68	48.14	7.15	98.51	51.25	51.98	43.32
LAQ2026.05	27.33	31.35	50.75	54.86	46.82	30.05	24.65	25.2	23.96	58.07	91.5	41.81	7.35	99.28	59.07	54.47	45.41
SnapKV2026.05	27.19	36.96	53.19	56.46	47.23	29.1	25.99	24.28	23.64	56.22	90.76	41.63	7.96	99.28	58.75	48.71	45.46
ZeroMerge2026.05	26.95	32.63	50.77	55.39	47.18	30.26	25.37	24.36	24.76	59.61	91.66	40.63	7.68	99.35	60.46	56.58	45.84
Judge Q2026.05	26.88	30.13	50.48	55.21	47.12	30.45	23.82	24.54	24.41	56.42	91.12	41.82	7.84	99.18	59.87	53.25	45.16
SnapKV2026.05	26.81	30.38	50.42	50.29	—	26.41	21.51	26.44	20.35	49.52	88.62	48.05	7.15	99.08	50.68	52.02	42.91
CaM2026.05	26.64	31.29	50.37	54.75	46.73	29.03	24.1	24.11	23.85	57.3	90.8	40.01	7.32	99.09	58.99	57.39	45.13
H2O2026.05	26.58	31.09	42.48	51.81	45.42	29.36	25.18	24.75	23.45	52.99	91.66	42.99	8.43	99.27	59.56	49.26	44.02
LAQ2026.05	26.52	28.19	49.36	54.12	46.12	29.54	22.45	25.12	23.12	54.85	91.24	41.52	7.12	99.12	58.42	54.47	44.46
H2O2026.05	26.4	23.96	36.53	44.67	37.02	27.16	21.75	27.3	20.65	48.85	89.65	49.82	7.14	98.98	52.07	51.73	41.48
ZeroMerge2026.05	26.19	29.88	49.45	54.82	46.52	29.82	23.41	24.12	24.12	56.84	91.45	40.12	7.52	99.21	60.12	56.57	45.01
CaM2026.05	25.84	28.45	49.12	54.12	46.12	28.42	22.12	23.84	23.12	54.41	90.45	39.45	7.12	98.92	58.42	57.37	44.21
SnapKV2026.05	25.12	31.62	50.71	55.45	45.58	27.21	20.81	23.54	21.21	46.51	89.65	40.71	7.68	98.85	56.81	46.51	43
StreamingLLM2026.05	24.89	24.18	35.89	42.39	34.08	25.27	22.04	24.89	19.79	49.97	83.46	46.39	7.15	98.04	49.26	49.35	39.51
H2O2026.05	24.82	24.81	36.54	49.24	43.52	28.12	21.15	24.42	21.65	44.82	91.24	42.92	8.41	98.85	58.51	47.92	42.35
StreamingLLM2026.05	23.48	25.18	35.89	46.7	40.04	26.21	21.44	22.17	20.71	46.02	85	39.5	7.5	98.33	55.4	45.72	39.95
StreamingLLM2026.05	22.92	19.95	31.35	39.87	31.95	23.67	18.88	23.56	17.94	43.97	81.54	45.15	7.15	97.54	47.63	47.41	37.53
StreamingLLM2026.05	22.01	21.14	31.82	44.51	38.12	25.02	18.84	21.52	19.34	40.82	83.54	38.81	7.32	98	54.12	44.42	38.08