Iambic released Enchant v3, the third generation of the multimodal transformer behind its drug discovery platform. Iambic states the model has 41 billion parameters, was pretrained on 4.5 trillion tokens, and covers more than 6,000 molecular properties across 16 biomedical data modalities using a mixture of experts architecture. Iambic says it will be used on internal and partner programs.
Our readEnchant v3 is the model partners get when they work with Iambic, so its prediction accuracy is the thing being bought. Iambic says scaling has held across three generations, including on endpoints with sparse data. Ask for the benchmark results in the Enchant v3 report and how they hold on targets with little training data.