Long-context Retrieval on NIAH 64k benchmark leaderboard · SOTA2 Research

Long-context Retrieval on NIAH 64k

49.3Single ScoreOriginal

Evaluation Results

Method	Links
Original2026.04		49.3	44.7	47
Original2026.04		49.3	45.7	47.5
Original2026.04		49.1	45.9	47.5
Original2026.04		47.9	23.5	35.7
BOSCH2026.04		43.7	23.6	33.7
BOSCH2026.04		41.3	13.9	27.6
BOSCH2026.04		38.5	11.1	24.8
FISHER2026.04		37	8.7	22.9
FISHER2026.04		37	13.3	25.2
BOSCH2026.04		32	15	23.5
FISHER2026.04		27.7	7.5	17.6
FISHER2026.04		23.9	5.7	14.8
B-layer2026.04		23.7	8.2	16
B-layer2026.04		19.5	3.8	11.7
B-layer2026.04		18.1	1.6	9.9
INTR2026.04		17.3	2.5	9.9
B-layer2026.04		12.1	2.7	7.4
INTR2026.04		2.9	1.8	2.4
INTR2026.04		2.6	1.7	2.2
INTR2026.04		2.6	1.3	2