OpenAI has released GPT-Rosalind, its first domain-specific model series, built to accelerate biology, drug discovery, and translational medicine research. Announced on April 16 and rolling out through this week, the model marks OpenAI's most explicit move yet from general-purpose chat assistants toward narrow, expert-grade scientific tools.
The model is named after Rosalind Franklin, the crystallographer whose X-ray diffraction work helped uncover the structure of DNA. OpenAI says GPT-Rosalind is fine-tuned for chemistry, protein engineering, and genomics, and is intended to help researchers move faster through the most time-intensive analytical stages of their work — not to replace scientists.
What it does
GPT-Rosalind synthesizes scientific evidence, generates hypotheses, plans experiments, and links into more than 50 scientific databases and computational tools through a new Life Sciences research plugin for Codex. Tasks the model is positioned for include literature review, protein structure analysis, cloning protocol design, and RNA sequence prediction, all within a single interface.
Benchmarks and partners
OpenAI reports a 0.751 pass rate on BixBench, a bioinformatics benchmark developed by FutureHouse and ScienceMachine that evaluates models on real-world computational biology tasks. The company says GPT-Rosalind outperformed GPT-5.4 on 6 of 11 LABBench2 tasks. In an evaluation run with Dyno Therapeutics inside the Codex app, best-of-ten model submissions ranked above the 95th percentile of human experts on the prediction task and around the 84th percentile on sequence generation.
Launch partners named in the announcement include Amgen, Moderna, Thermo Fisher Scientific, the Allen Institute, and Los Alamos National Laboratory — a mix of pharma, instrumentation, and federally backed research that signals OpenAI is targeting both commercial drug discovery pipelines and basic research.
Trusted-access only
GPT-Rosalind is available through ChatGPT, Codex, and the OpenAI API, but only to qualified enterprise customers in the United States through a gated trusted-access program. Organizations must focus on improving human health, conduct legitimate life sciences research, and maintain strong security governance to qualify. The restrictions echo the approach OpenAI has taken with cybersecurity-adjacent models: domain-specific capability gated by domain-specific oversight.
Why it matters
Life sciences has become the most aggressively contested vertical in frontier AI, with Anthropic acquiring Coefficient Bio earlier this month and Novo Nordisk signing a wide-ranging OpenAI partnership earlier this week. GPT-Rosalind sharpens the pattern: instead of asking general models to handle scientific workflows as one task among many, labs are stacking specialized models on top of trusted general ones. If the BixBench and LABBench2 numbers hold up under independent review, GPT-Rosalind will pressure both Google DeepMind's Isomorphic Labs efforts and the open-weight bio models from Meta and Alibaba to publish comparable evaluations. The trusted-access gating, meanwhile, gives OpenAI a defensible position on biosafety as regulators continue to scrutinize what frontier models can do with synthesis-grade chemistry and genomics data.
