Tamarind Bio Benchmarks brings together comparisons of more than ten widely used co-folding methods across curated molecular targets.
- Browse leaderboards for antibody-antigen, protein-ligand, and protein-protein benchmarks, including curated sets from FoldBench and Cofoldarena.
- Inspect individual predicted biomolecules against their experimental structures.
- Compare two structure-prediction methods across the same benchmark set.
- Use the model router to select a method for a specific input.
A benchmark winner is only a guide for a particular task. Model performance can shift with the input, which is why the page pairs average results with an input-aware router.
The benchmark collection also includes structure predictions for PDB entries outside model training data, offering comparisons on examples closer to new research inputs.
More model types and modalities are planned for the benchmark and router as the collection grows.