Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Long-context Retrieval on NIAH

100Accuracy

Vanilla

32.098449.726767.35584.9833Nov 3, 2025Dec 12, 2025Jan 21, 2026Mar 2, 2026Apr 11, 2026May 21, 2026Jun 30, 2026
Updated 24d ago

Evaluation Results

MethodLinks
2025.11
100
2025.11
100
2025.11
100
2025.11
100
2025.11
100
2026.06
95.28
2026.06
91.52
2026.06
91.17
2026.06
87.42
2026.06
85.23
2026.06
84.83
2026.06
83.17
2026.06
82.13
2026.06
81.72
2026.06
81.34
2026.06
80.74
2026.06
79.23
2026.06
78.41
2026.06
76.83
2026.06
76.72
2026.06
75.83
2026.06
75.47
2026.06
73.83
2026.06
73.42
2026.06
72.39
2026.06
72.14
2026.06
68.83
2026.06
68.72
2026.06
68.47
2026.06
66.83
2026.06
65.74
2026.06
64.83
2026.06
63.42
2026.06
61.17
2026.06
60.23
2026.06
59.84
2026.06
58.41
2026.06
58.34
2026.06
58.14
2026.06
56.83
2026.06
56.43
2026.06
54.17
2026.06
54.17
2026.06
53.17
2026.06
51.34
2026.06
47.83
2026.06
47.82
2026.06
43.72
2026.06
38.43
2026.06
34.71