<p>OpenAI has released GPT-Rosalind, its first domain-specific model series aimed at life sciences research. The name is a nod to crystallographer Rosalind Franklin.</p>
<p>The model is built to handle biochemistry, genomics, and protein engineering workflows — things like evidence synthesis, hypothesis generation, and experimental planning. It’s not a general-purpose upgrade; it’s scoped specifically to research contexts in biology, drug discovery, and translational medicine.</p>
<h2>Access is restricted</h2>
<p>OpenAI isn’t making this one widely available. Access goes through a vetted trusted-access programme for enterprise customers, with Amgen, Moderna, and Thermo Fisher Scientific among the initial partners. The restriction is deliberate — OpenAI cited dual-use biosecurity concerns and requires partner organizations to demonstrate they’re working toward improving human health outcomes.</p>
<p>A companion Life Sciences plugin for Codex connects the model to over 50 scientific tools and databases.</p>
<h2>Benchmark results</h2>
<p>On BixBench, GPT-Rosalind achieved a 0.751 pass rate and outperformed GPT-5.4 on six of eleven LABBench2 tasks. A third-party evaluation by Dyno Therapeutics found it ranking above the 95th percentile of human experts on sequence-to-function prediction — which is either impressive or a little unsettling depending on how you think about these things.</p>
<p>The restricted rollout suggests OpenAI is at least aware that a model this capable in biology warrants some caution before broader release.</p>


Collections

OpenAI has launched GPT-Rosalind, a domain-specific AI model series targeting life sciences research including biochemistry, genomics, protein engineering, drug discovery, and translational medicine. Access is restricted to vetted enterprise partners such as Amgen, Moderna, and Thermo Fisher Scientific due to dual-use biosecurity concerns. The model connects to over 50 scientific tools via a Life Sciences plugin for Codex. Benchmark results show a 0.751 pass rate on BixBench, outperforming GPT-5.4 on six of eleven LABBench2 tasks, and ranking above the 95th percentile of human experts on sequence-to-function prediction in a third-party evaluation by Dyno Therapeutics.

OpenAI launches GPT-Rosalind, a life sciences AI model with restricted access