OpenAI GPT-Rosalind AI model to speed up drug discovery |
Getting a drug from initial target discovery to U.S. regulatory approval takes between 10 and 15 years on average. Most of that time isn't spent on breakthroughs it's spent parsing literature, querying databases, and interpreting ambiguous results. On Thursday, OpenAI launched a model built specifically to attack that grind.
What OpenAI GPT-Rosalind is and who it's named after?
GPT-Rosalind represents OpenAI’s first attempt at building a domain-specific reasoning model specifically for biology, drug discovery, and translational medicine. This is the first model released in what OpenAI terms its Life Sciences model line-up and the first time the company has named one of its models after an individual from history.
The model is represented by a person who has the name Rosalind Franklin. Rosalind Franklin, a British chemist, contributed to the discovery of DNA's double-helix structure by using X-ray crystallography in the 1950s. Franklin was not credited with her findings while she was alive. The naming decision shows how OpenAI treats its newly entered field because it operates with a serious approach to research.
In a benchmark test designed to assess models’ abilities to perform real-world bioinformatics tasks, GPT-Rosalind scored an impressive 0.751, higher than any other model published to date on BixBench. GPT-Rosalind also outperformed its predecessor GPT-5.4 in six of eleven tasks in LABBench2.
In sequence prediction work conducted with gene therapy company Dyno Therapeutics using unpublished RNA sequences specifically to rule out memorisation GPT-Rosalind's best-of-ten submissions ranked above the 95th percentile of human experts, and around the 84th percentile on sequence generation tasks.
OpenAI's own life sciences research lead, Joy Jiao, was careful not to overstate what those numbers mean in practice. The model is not designed to develop new treatments autonomously, she said at a press briefing. The goal is speed, not replacement. "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 told reporters.
GPT-Rosalind is not publicly available. Access is currently restricted to US enterprise customers who pass a qualification and safety review a deliberate response to an open letter from more than 100 scientists calling for tighter controls on biological data used to train AI, citing the risk of pathogen design misuse.
Launch partners include Amgen, Moderna, and Thermo Fisher Scientific. OpenAI is also running a separate research collaboration with Los Alamos National Laboratory focused on AI-guided protein and catalyst design.
Alongside the model, the company is releasing a free Life Sciences research plugin for its Codex platform, connecting users to over 50 scientific databases covering protein structure, genomics pipelines, sequence search, and literature review. Enterprise GPT-Rosalind users receive the full reasoning layer on top; everyone else gets the plugin with standard models.
No AI-discovered drug has cleared Phase 3 clinical trials. That number is still zero. But if GPT-Rosalind compresses early-stage research cycles across thousands of labs simultaneously, the compounding effect on what gets discovered and when may ultimately matter more than any single benchmark score.