RareBench
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
RareBench RAMEDIS
73.23Recall@1
21
RareBench LIRICAL
55.95R@1
21
RareBench HMS
56.97Recall@1
21
RareBench MME
77.5R@1
21
RareBench Dirichlet α=0.3 partition
68Accuracy
18
RareBench (test)
97.5Property Score
15
RareBench v1 (test)
96.9Property Alignment
11
RareBench
96.3Property (Single Object)
11
RareBench-Public Phenotypes
37.92Accuracy
9
RareBench HMS subset n=88
51.14Recall@1
7
RareBench MME n=40
40R@1
7
RareBench (combined)
39.31Recall@1
7
RareBench LIRICAL
36.24Recall@1
7
RareBench Qwen-0.8B text (test)
85.6Hit@1
5
RareBench
7.36Latency (sec)
4
RareBench
85.6Conditional Hit@1
4
RareBench
46.79Pass@1
4
RareBench Overall
74.5Pick Score
3
RareBench Advanced
71.5Pick
3
RareBench Basic
95.5Pick Score
3
RareBench
76.7Property Score
3
RareBench (full)
6.395LAION-aesthetic
3
RareBench 1.0 (test)
81.9Property Score
3
RareBench
—Primary metric
0