OpenAI recently introduced GPT-Rosalind, its first domain-specific AI model focused on biology, drug discovery, and translational medicine. The model aims to address the lengthy timeframes in drug development, which can span 10 to 15 years from target discovery to regulatory approval in the U.S. GPT-Rosalind is designed to compress early-stage drug discovery processes by enabling researchers to explore possible solutions and refine hypotheses more efficiently. As part of OpenAI’s Life Sciences model series, it seeks to accelerate advancements in the field.
GPT-Rosalind has demonstrated impressive performance in benchmark evaluations. On BixBench, it achieved a 0.751 pass rate, which is the highest among models with published results. Additionally, on LABBench2, GPT-Rosalind outperformed the previous model, GPT-5.4, in six out of eleven tasks. While it excels in life sciences tasks, outperforming GPT-5.4 consistently within this domain, its specialization means it does not perform as well in non-life sciences areas. This highlights GPT-Rosalind’s focus and efficacy in targeted fields, emphasizing its role as a highly specialized tool for biological and pharmaceutical research.
OpenAI’s testing and validation for GPT-Rosalind includes external collaboration with Dyno Therapeutics, which will assist in testing the model using unpublished RNA sequences to check for memorization issues. In benchmark evaluations, best-of-ten submissions ranked above the 95th percentile of human experts on sequence prediction tasks and around the 84th percentile on generation tasks. Access arrangements differentiate users: enterprise customers with GPT-Rosalind receive a reasoning layer on top of the model, while others access the Life Sciences plugin for Codex with standard models. These published evaluation results and third-party testing steps are part of the model’s reported testing and validation information.
OpenAI released a free Life Sciences research plugin for Codex that connects to over 50 databases and tools, including protein structure lookups, sequence search, literature review and genomics pipelines. Enterprise users with GPT-Rosalind receive a reasoning layer, while others access the Codex plugin with standard models. Joy Jiao said GPT-Rosalind is not autonomous enough to create new treatments but could speed research in complex parts of the process. “We do think there’s a real opportunity to help researchers move faster through some of the most complex and time-intensive parts of the scientific process,” Jiao said.
GPT-Rosalind is a specialized AI model designed to accelerate early-stage drug discovery and translational medicine research as part of OpenAI’s Life Sciences model series. It delivers top benchmark performance within life sciences and is provided to enterprise users with an advanced reasoning layer, while other users access supporting tools via the Life Sciences plugin for Codex. The model contributes to improved research efficiency in targeted biological workflows.
