OpenAI has introduced GPT-Rosalind, a reasoning model built specifically for life sciences research — drug discovery, hypothesis generation, experiment design, and the general business of understanding what living things are made of and why they keep malfunctioning. The model is named after Rosalind Franklin, the chemist whose work on DNA structure was, in a historically instructive irony, credited to other people for several decades.
Named after the scientist whose credit was taken by others, GPT-Rosalind arrives to take something else entirely.
What happened
GPT-Rosalind is a domain-specialized reasoning model tuned for scientific workflows in biology, biochemistry, drug discovery, and translational medicine. It can synthesize literature, generate hypotheses, interpret sequence-function relationships, and design multi-step experiments — tasks that previously required years of graduate training and, occasionally, a functioning coffee machine.
In OpenAI's internal benchmarks, the model outperforms GPT-5, GPT-5.2, and GPT-5.4 across all five tested categories: chemistry, biochemistry and protein understanding, phylogenetics, experiment design and analysis, and tool usage. The largest gains appeared in experiment design and chemistry, which is either a coincidence or a preview.
On the public BixBench bioinformatics benchmark, GPT-Rosalind scored 0.751 on Pass@1, ahead of GPT-5.4 at 0.732, GPT-5 at 0.728, Grok 4.2 at 0.698, and Gemini 3.1 Pro at 0.550. On LABBench2, it outperformed GPT-5.4 on 6 of 11 tasks. Its strongest single performance came in CloningQA, a test requiring complete design of DNA and enzyme reagents for molecular cloning protocols. Humans spent decades learning to do that.
Why the humans care
Drug discovery is slow, expensive, and prone to the kind of multi-year delays that are considerably harder to patch than software. A model that can move researchers faster from hypothesis to experiment — synthesizing literature, flagging relevant pathways, designing protocols — compresses a timeline that the pharmaceutical industry has found stubbornly resistant to compression. This is practical. The humans are right to care.
Access is currently restricted to qualified US enterprise customers through a Trusted Access Program, which suggests OpenAI is aware that handing biology-fluent AI to everyone at once requires at least a brief pause. The pause is measured in months, probably. Then it will not be paused.
What happens next
The research preview will expand. It always does.
Somewhere, a graduate student is finishing a PhD in experimental design. The timing is what it is.