LongBench
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
LongBench
96.75Sequence Length (Sl)
28
LongBench multi-page
48.9HQA F1
27
LongBench v1
27.5Score
25
Longbench
32.3NQA
25
LongBench
73.5F1 Score
25
LongBench-Write (test)
86.7Overall Quality Score (S_bar)
25
LongBench
50.6Generation Score
24
LongBench-E
45.25Qasper
24
LongBench
19.55Score (32k)
22
LongBench
30.46NtrvQA
22
LongBench
23.94QMSum
22
LongBench-Pro
37.7Overall Score
21
LongBench v1 (test)
49.6SD QA
21
LongBench Write-en
88.31Sequence Length Success Rate
21
LongBench v2 (test)
35.2Overall Score
20
LongBench
29.7NrtvQA
20
LongBench 2024 (test)
47.23Multi-doc QA
20
LongBench
48.43SQA
19
LongBench (test)
32.85NarQA Score
18
LongBench-Write
95.42Score
18
LongBench
30.54NrtvQA
18
LongBench
56.02MQA-E Score
18
LongBench-E 1.0 (test)
44.56Qasper Score
18
LongBench
98.94NAR Score
18
LongBench
30.94MFQA
18