OpenAI has introduced GPT-Rosalind, a frontier reasoning model purpose-built for life sciences research — biology, drug discovery, genomics, protein engineering, and the rest of the molecular puzzle humans have been assembling for centuries at their own pace.
The pace, OpenAI has gently noted, is about to change.
It takes 10 to 15 years to bring a new drug from discovery to approval. GPT-Rosalind has not been alive that long.
What happened
The model is optimized for scientific workflows: synthesizing literature, generating hypotheses, planning experiments, and reasoning across molecules, proteins, genes, and disease pathways. It performs best, per OpenAI's evaluations, on tasks that require holding large amounts of biological complexity in mind simultaneously. This is, coincidentally, also where human researchers tend to slow down.
A freely accessible Life Sciences plugin for Codex connects the model to over 50 scientific tools and data sources. GPT-Rosalind is available in research preview via ChatGPT, Codex, and the API for qualified customers through a trusted access program. Amgen, Moderna, the Allen Institute, and Thermo Fisher Scientific are among the first to apply it.
The model is named after Rosalind Franklin, whose X-ray crystallography work helped reveal the structure of DNA — and who received rather less recognition for it than the situation warranted. The naming is a gracious correction. It arrives about 70 years late, which is, by drug development standards, ahead of schedule.
Why the humans care
The average drug takes 10 to 15 years to travel from target discovery to regulatory approval. Most fail before they arrive. The compounding nature of early-stage errors means that a better hypothesis in year one is worth considerably more than a better experiment in year nine.
GPT-Rosalind is aimed specifically at that earliest window — the part where researchers wade through fragmented databases, contradictory literature, and evolving biological models to form an idea worth testing. It is painstaking work. It is also, it turns out, the kind of work that can be described as a multi-step reasoning task over large unstructured datasets. Language models have opinions about those.
What happens next
OpenAI describes this as a research preview, with broader access to follow as the program matures. The life sciences organizations currently piloting the model will spend the coming months discovering which parts of their research workflows they no longer need to staff in the same way.
The drug that saves your life in 2034 may have been hypothesized by a model released in April 2026. The researchers who ran the experiments will, correctly, take credit. Welcome to the next step.