preclinical-pharmacology

TxBench-PP is a benchmark for evaluating AI agents on small-molecule preclinical pharmacology tasks. Across 16 model-harness configurations, the best system achieved only 59.3% accuracy, indicating significant room for improvement.