Materials Stability Prediction on Matbench-Discovery
92.9F1 ScoreUMA-M-1.1
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| UMA-M-1.1fine-tuned=MPtrj [17] and sAlex [5, 62]2025.06 | 92.9 | 0.061 | 0.018 | 0.176 | — | — | — | — | — | |
| eSEN-30M-OAM2025.06 | 92.5 | 0.061 | 0.018 | 0.17 | — | — | — | — | — | |
| UMA-S-1.2fine-tuned=MPtrj [17] and sAlex [5, 62]2025.06 | 92.1 | 0.062 | 0.019 | 0.129 | — | — | — | — | — | |
| UMA-S-1.1fine-tuned=MPtrj [17] and sAlex [5, 62]2025.06 | 91.3 | 0.064 | 0.02 | 0.204 | — | — | — | — | — | |
| ORB v32025.06 | 90.5 | 0.075 | 0.024 | 0.21 | — | — | — | — | — | |
| SevenNet-MF-ompa2025.06 | 90.1 | 0.064 | 0.021 | 0.317 | — | — | — | — | — | |
| GRACE-2L-OAM2025.06 | 88 | 0.067 | 0.023 | 0.294 | — | — | — | — | — | |
| MACE-MPA-02025.06 | 85.2 | 0.073 | 0.028 | 0.412 | — | — | — | — | — | |
| Pruned MACE-MPModel=Ours, Architecture=Layer2 L0 C128, Training Data=25% of MPtrj2026.05 | 68 | 0.0917 | 0.056 | 0.625 | 3.889 | 59.5 | 79.4 | 88.4 | 0.717 | |
| MACE-MP-0Model=Official Checkpoint, Architecture=Layer2 L1 C1282026.05 | 66.9 | 0.0915 | 0.057 | 0.682 | 3.777 | 57.7 | 79.6 | 87.8 | 0.697 |