ORD Reaction Benchmark

Reaction benchmarks derived from the Open Reaction Database (yield prediction, condition recommendation).

Composite
93.9
Experimental validation
Retrospective
Stages
Developmental Candidate
Modalities
small-molecule
Task types
reaction-predictionyield
Size
reactions: 2,100,000
tasks: 4
License
CC-BY-SA 4.0
First release
2021-07
Last updated
2025-04
Official site
→ project page
Leaderboard
→ leaderboard
Dataset
→ dataset
Code / GitHub
→ repository
HuggingFace
→ HF
Paper
The Open Reaction Database · Kearnes SM, Maser MR, Wleklinski M, et al. · 2021 · paper · doi:10.1021/jacs.1c09820 · 210 citations
Flags
none
Experts
Steven Kearnes, Abigail Doyle, Connor Coley
Groups
Google Research (Kearnes), Princeton Chemistry (Doyle, ex), Coley Lab (MIT)
Hosted by
Open Reaction Database (ORD)
Related benchmarks
USPTO-50K / USPTO-MIT (Retrosynthesis)

Rubric (7-criterion)

rigor
5
coverage
4
maintenance
5
adoption
4
quality
5
accessibility
5
industry_relevance
5

Notes

Modern open reaction corpus; industry-scale.

← Back to all benchmarks

Compare:
Open comparison →