LiteFold benchmarks

LiteMol-1

LiteMol-1 is LiteFold's first molecular foundation model for agent-driven design. The benchmark panel summarizes peptide co-fold confidence across ipSAE, ipTM, and pLDDT, plus small-molecule docking strength against public baselines.

SystemScore (score)
LiteMol-1 + MCTS8.02
ProtoBind-Diff8.17
LiteMol-17.77
PocketXMol7.18

MitoTox

MitoTox is LiteFold's state-of-the-art model for mitochondrial toxicity prediction, part of our pre-clinical stack (Signal). It helps teams screen liabilities early and run toxicity-aware lead design and optimization.

SystemScore (%)
MitoTox (LiteFold)82.6
Atom-Pair + GB Baseline80.9
Mammoth (LiteFold)39.5

Molecular Dynamics

STMV benchmark (1.07M atoms, explicit PME), higher is better. Our optimized engine brings datacenter-class molecular dynamics throughput to a single RTX 5090, measured in nanoseconds simulated per day. Representative benchmark comparing our engine against stock OpenMM and datacenter GPUs.

SystemScore (ns/day)
RTX 5090 (LiteFold)136
B200118
RTX 5090 (Stock)101
H20078
H10062
RTX 408046
A10032

BixBench

BixBench is the Future House benchmark of 200+ real bioinformatics tasks built from published research notebooks. Agents have to navigate complex datasets, execute Python and R, generate testable hypotheses, and defend their answers. New re-evaluations are ongoing right now, and the leaderboard will be refreshed as results finish.

SystemScore (%)
Rosalind (LiteFold)90.24
K-Dense90.0
Biomni Lab88.7
Edison78.0
Claude Code (Opus 4.6)65.3