OpenAI has introduced its first domain-specific artificial intelligence model named GPT-Rosalind, in honor of British chemist Rosalind Franklin, known for her pivotal work on DNA’s double helix structure. Revealed this Thursday, GPT-Rosalind is crafted specifically to enhance reasoning capabilities in biology, drug discovery, and translational medicine. This initiative marks the beginning of OpenAI’s Life Sciences model series, competing directly with specialized labs from universities to Google DeepMind.
Experts note that bringing a new drug from target identification to U.S. regulatory approval typically spans 10 to 15 years. Much of this duration is consumed not by breakthroughs but by meticulous tasks: reviewing thousands of papers, querying databases, designing reagents, and interpreting ambiguous data—tasks GPT-Rosalind aims to streamline.
OpenAI posits that the model can significantly condense these early-stage efforts. The company emphasizes that GPT-Rosalind is engineered to enable scientists to explore more possibilities, identify connections easily overlooked, and formulate superior hypotheses more quickly.
The model’s efficacy is supported by benchmarks: it achieved a 0.751 pass rate on BixBench, surpassing other models with published outcomes in real-world bioinformatics tasks. Additionally, GPT-Rosalind outperformed its predecessor GPT-5.4 across six of eleven tasks on LABBench2.
Despite excelling in life sciences-related tasks, GPT-Rosalind is a highly specialized model that may underperform outside this domain. To evaluate the model’s capabilities further, OpenAI has partnered with Dyno Therapeutics to test it using unpublished RNA sequences, ensuring it doesn’t simply memorize data. On sequence prediction tasks, its top submissions ranked above the 95th percentile of human experts, and around the 84th percentile for generation.
However, Joy Jiao, OpenAI’s life sciences research lead, tempers expectations. She clarified that while Rosalind isn’t designed to autonomously create new treatments, it can significantly expedite scientific research processes by helping researchers navigate complex and time-intensive stages more efficiently. “We believe there is a substantial opportunity for accelerating the pace of research in challenging segments of the scientific process,” Jiao stated during a press briefing reported by theLA Times.
The model’s ecosystem could be as crucial as the technology itself. OpenAI plans to release a free Life Sciences research plugin for Codex, linking over 50 databases and tools including protein structure lookups, sequence searches, literature reviews, and genomics pipelines. Enterprise users with GPT-Rosalind access will gain enhanced reasoning capabilities, while others can use the plugin with standard models.
OpenAI has secured partnerships with several pharmaceutical and biotech firms for its launch, such as Amgen, Moderna, and Thermo Fisher Scientific. Additionally, a research collaboration is underway with Los Alamos National Laboratory focusing on AI-driven protein and catalyst design.
“The life sciences industry requires precision at every step due to the complexity of questions, uniqueness of data, and high stakes involved,” stated Sean Bruich, Amgen’s Senior VP of AI and Data. Access to Rosalind is intentionally restricted; it is U.S.-enterprise only, requiring qualification and safety reviews. This cautious approach responds to concerns raised by an international coalition of scientists about the risks associated with training AI on biological data.
OpenAI isn’t new to integrating science workflows into its offerings, having launched the Prism scientific writing workspace in January as a precursor. GPT-Rosalind represents a more focused and specialized advancement, signaling that domain-specific models are increasingly becoming a competitive edge.
To date, no drug discovered entirely by AI has completed phase 3 trials. Nevertheless, if GPT-Rosalind enables researchers to design improved experiments six months earlier across numerous labs, the cumulative impact on discoveries could be transformative. This potential effect is central to OpenAI’s strategy and merits close observation.