SupraBench: A Benchmark for Supramolecular Chemistry

The article announces the release of SupraBench, a benchmark specifically designed for evaluating large language models (LLMs) on tasks related to supramolecular chemistry, including binding affinity prediction and host-guest description. It introduces four fundamental tasks alongside an auxiliary vision-based task, supported by SupraPMC, a 16M-token corpus of relevant articles for domain adaptation. This benchmark highlights significant performance gaps in current LLMs, providing a structured framework for researchers to enhance model capabilities in supramolecular chemistry reasoning.