Chai Discovery's introduction of Chai-2 represents a significant advancement in molecular design, showcasing a major breakthrough in zero-shot de novo antibody design. The Chai-2 series of models has achieved a remarkable double-digit success rate, a vast improvement over previous methods that typically required screening millions to billions of candidates to find a single viable hit. By contrast, Chai-2 can routinely discover successful binders by testing as few as 20 designs in a 24-well plate format, compressing the discovery timeline from months or years down to just two weeks.

At the core of this innovation is a multimodal generative AI architecture that integrates all-atom structure prediction with generative modeling. This allows Chai-2 to design novel and diverse binders—including scFv antibodies, nanobodies, and miniproteins—that are highly specific to their target epitope. In a recent validation, the model achieved a 16% binding rate across 52 diverse protein targets, successfully finding at least one binder for half of them. The designed antibodies were also found to be stable and specific, with very little similarity to existing antibodies.

This paradigm shift from stochastic screening to intentional, programmable design has profound implications for drug discovery. It opens the door to developing therapies for previously "undruggable" targets and could significantly streamline the creation of advanced therapeutic formats like antibody-drug conjugates. Chai Discovery believes these results mark a key milestone in its mission to transform biology from a science of discovery into a discipline of deterministic engineering. The company is now offering access to Chai-2 to select academic and industry partners to help accelerate research with a positive impact on human health.