OpenAI is going a different route: instead of building one model for everything, there’s now GPT-Rosalind — the first OpenAI model built exclusively for biology, genomics, and drug discovery. Named after Rosalind Franklin, whose work helped unlock the structure of DNA.
What GPT-Rosalind Can Do
The model was trained on the 50 most common biological workflows and has access to major public biological databases. It can synthesize evidence from multiple sources, generate biological hypotheses, and plan experiments — tasks that used to require years of expert knowledge.
The benchmarks look promising: on BixBench, a bioinformatics benchmark, Rosalind achieves a pass rate of 0.751. On LABBench2, which measures research tasks like literature retrieval and protocol design, it outperforms GPT-5.4 on 6 out of 11 tasks.
Who Gets Access
This is where it gets interesting: GPT-Rosalind isn’t available to everyone. OpenAI restricts access to organizations working on improving human health outcomes with strong security and governance controls. Initial partners include Amgen, Moderna, the Allen Institute, and Thermo Fisher Scientific.
The model is available as a research preview in ChatGPT, Codex, and the API — but only through OpenAI’s Trusted Access Program.
My Take
This is a smart move. Instead of trying to build a model that’s good at everything, OpenAI is going vertical. And the restricted access isn’t a marketing gimmick — for a model that can generate hypotheses about disease mechanisms, caution makes sense. The question is whether Anthropic is taking a similar path with Mythos for cybersecurity, and whether we’re about to see a whole wave of specialized frontier models. I think we are.
Sources: