Bandgap Prediction on Matbench Bandgap
0.091MAE (eV)JMP-L
Evaluation results
| Method | Links | |
|---|---|---|
| JMP-LPre-training strategy=FE Pretrained, Params=235M, Data=120M2026.03 | 0.091 | |
| JMP-SPre-training strategy=FE Pretrained, Params=27M, Data=120M2026.03 | 0.121 | |
| CT-SCD-AMP20Pre-training strategy=SCD Pretrained (Ours), Params=10M, Data=675k2026.03 | 0.123 | |
| CT-SCD-ALLPre-training strategy=SCD Pretrained (Ours), Params=10M, Data=11.3M2026.03 | 0.132 | |
| CT-FE-OMOL25Pre-training strategy=FE Pretrained, Params=10M, Data=4M2026.03 | 0.134 | |
| CT-SCD-OMOL25Pre-training strategy=SCD Pretrained (Ours), Params=10M, Data=4M2026.03 | 0.136 | |
| HackNIPPre-training strategy=FE Pretrained, Params=> 25M, Data=> 32M*2026.03 | 0.15 | |
| Elements2026.04 | 0.1514 | |
| coGN2026.04 | 0.1559 | |
| coGNPre-training strategy=No Pretraining, Params=—, Data=n/a2026.03 | 0.156 | |
| coNGN2026.04 | 0.1697 | |
| CT (Baseline)Pre-training strategy=No Pretraining, Params=10M, Data=n/a2026.03 | 0.186 | |
| ALIGNN2026.04 | 0.1861 | |
| MEGNet2026.04 | 0.1934 | |
| DimeNet++2026.04 | 0.1993 | |
| MODNet2026.04 | 0.2199 | |
| MODNetPre-training strategy=No Pretraining, Params=—, Data=n/a2026.03 | 0.22 | |
| JMP-LPre-training strategy=No Pretraining, Params=235M, Data=n/a2026.03 | 0.228 | |
| JMP-SPre-training strategy=No Pretraining, Params=25.2M, Data=n/a2026.03 | 0.235 | |
| SchNet2026.04 | 0.2352 | |
| CGCNN2026.04 | 0.2972 |