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

Long-context Reasoning Accuracy on LongBench

68.7Accuracy (LongBench)

AoT

52.0656.3860.765.02Mar 13, 2026
Updated 11d ago

Evaluation Results

MethodLinks
2026.03
68.7
2026.03
68.5
2026.03
68.1
2026.03
63.4
2026.03
63.2
2026.03
63.1
2026.03
62.8
2026.03
62.8
2026.03
62.4
2026.03
61.4
2026.03
61
2026.03
60.4
2026.03
59.5
2026.03
59.5
2026.03
59.3
2026.03
59.3
2026.03
59.3
2026.03
59.1
2026.03
59
2026.03
59
2026.03
59
2026.03
59
2026.03
58.9
2026.03
58.8
2026.03
58.7
2026.03
58.6
2026.03
58.6
2026.03
58.6
2026.03
58.5
2026.03
58.5
2026.03
58.4
2026.03
58.4
2026.03
58.4
2026.03
58.3
2026.03
58.2
2026.03
58.2
2026.03
58.1
2026.03
58
2026.03
58
2026.03
57.7
2026.03
57.6
2026.03
57.3
2026.03
53.4
2026.03
52.9
2026.03
52.7