AI with
chemical insight.

An open-source tool for naming molecules, built by a verifier-guided coding agent. NISPO achieved 98.1% round-trip accuracy across 103 million molecules.

An explicit molecular graph format for language models. Across 224,055 model responses, MolJSON substantially outperformed common formats on molecular translation, graph reasoning, and generation tasks.

The first systematic assessment of reasoning models in chemistry. Across 816 questions, ChemIQ showed that LLMs had gained the ability to perform chemical reasoning, including elucidating molecular structures from NMR data.

Inference-time conditioning of an equivariant diffusion model for fragment-based drug design, enabling fragment merging and linker generation without retraining.