Molecular Graph Generation on MOSES benchmark leaderboard · SOTA2 Research

Molecular Graph Generation on MOSES

100UniquenessTraining Set

Evaluation Results

Method
Training Set2026.01	100	100	0	100	0.01	0.64	99.1	—
DiGress2026.01	100	85.7	95	97.1	1.19	0.52	14.8	—
DisCo2026.01	100	88.3	97.7	95.6	1.44	0.5	15.1	—
JT-VAE2025.12	100	100	99.9	97.8	1	0.53	10	—
DiGress2025.12	100	85.7	95	97.1	1.19	0.52	14.8	—
PARD2025.12	100	86.8	78.2	99	1	0.56	2.2	—
GVT2025.12	100	99.42	61.43	99.59	0.16	0.64	3.4	—
DiGress2026.02	100	—	95	—	—	0.52	14.8	—
DisCo2026.02	100	—	97.7	—	—	0.5	15.1	—
G2PT2026.02	100	—	79.4	—	—	0.55	2.9	—
GEEL2026.02	100	—	81.1	—	—	0.52	3.6	—
PARD2026.06	100	86.8	78.2	—	1	—	—	—
Structure-Guided Autoregressive framework2026.06	100	95.25	98.72	—	19.2	—	—	—
Cometh2026.01	99.9	90.5	92.6	99.1	1.27	0.54	16	19,000
DeFoG2026.01	99.9	92.8	92.1	98.9	1.95	0.55	14.4	18,000
LG-Flow2026.01	99.9	88.4	90.5	98.9	1.43	0.55	12.3	310
ConGress2025.12	99.9	83.4	96.4	94.8	1.48	0.5	16.4	—
Cometh2026.02	99.9	—	92.6	—	—	0.54	16	—
DeFoG2026.02	99.9	—	92.1	—	—	0.55	14.4	—
Flowette2026.02	99.9	—	98.1	—	—	0.58	15.3	—
VAE2025.12	99.8	97.7	69.5	99.7	0.57	0.58	5.9	—
GraphINVENT2025.12	99.8	96.4	—	95	1.22	0.54	12.7	—
GraphBFN2026.02	99.8	—	89	—	—	0.59	10	—