Leash Bio · siet Juli 2025 (erstmals beschrieben am 17.07.2025 bzw. Substack-Beitrag vom 30.07.2025) · 5× · tolest 15. Okt. 2026
92
Momentum
Hermes is a compact transformer model from Leash Bio for predicting protein-ligand binding of small molecules. According to the company, it was trained exclusively on Leash's internal DEL binding data and has about 50M parameters, built on the pre-trained encoders ESM2 and ChemBERTa. Leash reports it is much faster than Boltz-2 and performs better on its own validation sets, with only a narrow gap on curated third-party benchmarks.
Momentum-Verloop
18.07.16.10.
Features
| Deployment Model | Not documented as a commercial hosted product; paper available on GitHub (Leash-Labs/hermes-paper) |
| Use Case Scope | Prediction of protein-ligand binding likelihood for small molecules; intended as fast first-pass filter before structure-based models |
| Platform | Transformer model (research model, published via Substack/GitHub paper) |
| Release Date | First described July 2025 |