Chai-1

Chai-1, released September 9, 2024 by Chai Discovery (founded by ex-OpenAI and Meta researchers, backed by OpenAI and Thrive Capital), is a multi-modal foundation model for molecular structure prediction. It folds proteins together with small-molecule ligands, glycans, DNA, and RNA, predicting joint 3D structures from sequence — and can run in single-sequence mode without multiple sequence alignments while preserving most of its performance. The company reports state-of-the-art results across drug-discovery-relevant tasks, with optional experimental restraints boosting performance by double-digit percentage points.

Limitations: the downloadable weights and inference code are released for non-commercial use; commercial drug-discovery use goes through the free web interface. As with all structure predictors, accuracy varies by target class and available evolutionary information.