Long-context reasoning on LongBench 256 tokens v2 benchmark leaderboard · SOTA2 Research

Long-context reasoning on LongBench 256 tokens v2

100AccuracyMamba

Evaluation Results

Method	Links
Mamba2026.02		100	21.99	96.86	100	100	83.77
Mamba2026.02		100	24.22	88.95	100	100	82.63
Mamba2026.02		100	26.55	95.63	100	100	84.44
Steered Mamba2026.02		100	36.61	99.73	100	100	87.27
Steered Mamba2026.02		100	38.33	91	100	100	85.87
Steered Mamba2026.02		100	40.42	95.02	100	100	87.09
Stable-Mamba2026.02		100	38.35	98.92	100	100	87.45
Stable-Mamba2026.02		100	42.14	91.17	100	100	86.66
Stable-Mamba2026.02		100	45.32	96.11	100	100	88.29
Hyena2026.02		100	14.85	89.78	100	100	80.93
Mamba-22026.02		100	16.71	92.82	100	100	81.91
MiniPLM-Mamba2026.02		100	13.55	87.44	100	100	80.2
GPT-22026.02		95	12.18	81.68	95	95	75.37
DenseMamba2026.02		80	16.63	82.41	80	100	71.81