OpenAI Debuts GPT-Rosalind Model for Drug Discovery

OpenAI has transitioned its specialized life sciences reasoning model, GPT-Rosalind, out of its research preview. The model is now available to qualified U.S. Enterprise customers through trusted access via the API, Codex, and ChatGPT Enterprise. Named after the pioneering scientist Rosalind Franklin, the model is designed to assist with multi-step tasks in early drug discovery, genomics, protein engineering, and translational medicine. During this initial preview phase, usage does not consume existing credits or tokens, though OpenAI has not yet announced final production pricing, rate limits, or context limits.

In evaluations, GPT-Rosalind demonstrated highly competitive capabilities. On a blind RNA sequence-to-function prediction task from Dyno Therapeutics, the model's best-of-10 submissions ranked above the 95th percentile of human AI-biology experts for prediction and near the 84th percentile for sequence generation. Additionally, the model outperformed GPT-5.4 on six out of 11 LABBench2 tasks and achieved the leading score among published models on the BixBench benchmark.

For practitioners, the release includes a free Life Sciences plugin for Codex that connects to more than 50 public multi-omics databases and scientific tools. This integration allows researchers to coordinate sequence searches, structure lookups, literature reviews, and dataset discovery across fragmented systems. While GPT-Rosalind remains gated by strict biosecurity and governance reviews, developers can test the orchestration layer and database connections using OpenAI's mainline models.

Early launch partners evaluating the model include major industry players and research institutions such as Amgen, Moderna, Novo Nordisk, Thermo Fisher Scientific, NVIDIA, Oracle Health and Life Sciences, the Allen Institute, Benchling, and the UCSF School of Pharmacy. By synthesizing evidence, planning experiments, and designing molecular cloning protocols, the model aims to streamline the historically slow 10-to-15-year drug development pipeline, though human experimental validation remains necessary.