Human Preference Alignment
Benchmarks
Dataset NameSOTA methodMetricTrendResultsLast Updated
0.324BLEU
31
Mar 23, 2026
6.83SAG
20
Jun 9, 2026
86.851PickScore
20
May 11, 2026
84.9Win Rate
18
Feb 26, 2026
63.9Win Rate (%)
16
Jun 9, 2026
32.06Anime Score
16
Apr 21, 2026
81.8Agreement
15
Mar 26, 2026
3.876Reward
14
Feb 26, 2026
61Acc (Tie, Overall)
12
Feb 26, 2026
0.379HPS-v2.1
10
Mar 19, 2026
0.371HPSv2.1
7
Mar 19, 2026
35.3HPS-v2.1
7
Feb 26, 2026
22.46Pick Score
7
Feb 26, 2026
37.7HPS-v2.1
6
May 13, 2026
0.361HPS-v2.1
6
May 13, 2026
0.36HPS-v2.1
6
May 13, 2026
49.4Avg@3
6
Feb 26, 2026
60Avg@3 Score
6
Feb 26, 2026
23.67PickScore (Training Reward)
5
May 12, 2026
23.64PickScore
5
Apr 7, 2026
0.374HPS-v2.1
5
Feb 26, 2026
23.55PickScore (Task)
5
Feb 26, 2026
66VQ Acc (w/ Tie)
5
Feb 26, 2026
3.93Quality Score
4
Mar 23, 2026
24.64PickScore (Task Metric)
4
Feb 26, 2026