OpenAI has released GPT-Rosalind — a specialized model for accelerating drug development

The name is given in honor of Rosalind Franklin, whose work helped reveal the structure of DNA and laid the foundation for molecular biology.

Why a separate model is needed

In the US, the path from discovering a molecular target to regulatory approval of a new drug takes an average of 10 to 15 years. OpenAI believes that progress in the early stages creates a cumulative effect on all subsequent stages — if scientists more accurately select targets, formulate hypotheses faster, and conduct higher-quality experiments.

The key problem that GPT-Rosalind solves is not the complexity of science itself, but the complexity of research processes. Scientists work with vast amounts of literature, specialized databases, experimental data, and constantly changing hypotheses.

What the model can do

GPT-Rosalind shows the best results in tasks related to reasoning about molecules, proteins, genes, biological pathways, and disease biology. It works more effectively with scientific tools and databases in multi-step processes: literature review, interpreting "sequence-function" relationships, experimental planning, and data analysis.

In open tests, the model demonstrates high results. In the BixBench test, based on real bioinformatics tasks, GPT-Rosalind showed the best results among all models with published metrics.

In the LABBench2 test, it surpasses GPT-5.4 in 6 out of 11 categories — the most significant improvement was recorded in tasks for complete design of DNA sequences for molecular cloning protocols.

In partnership with Dyno Therapeutics, specializing in gene therapy, the model was tested on tasks of predicting and generating RNA sequences. The best of the 10 proposed variants exceeded 95% of experts in the prediction task and about 84% in the sequence generation task.

Plugin and tools

Along with the model, OpenAI released a research plugin for life sciences — the Life Sciences research plugin for Codex. It is available for free on GitHub and provides access to more than 50 public databases, literature sources, and biological tools. Among the supported areas are human genetics, functional genomics, protein structure, biochemistry, and clinical data.

Enterprise-level users can use the plugin together with GPT-Rosalind for deeper analysis. For others, the plugin is available when working with basic OpenAI models.

Who has access

In the initial stage, GPT-Rosalind is launched as part of a trusted access program for qualified Enterprise customers in the US. Organizations must conduct scientific research with clear public benefit, maintain compliance control mechanisms, and restrict access only to approved users within secure environments.

During the preliminary research period, using the model does not consume existing credits or tokens. Access can be requested on the OpenAI website.