OpenAI's GPT-Rosalind Enters Drug Discovery Arena, But Only American Enterprises Need Apply CATEGORY: Industry News

OpenAI dropped a new reasoning model called GPT-Rosalind on Thursday—built specifically for biology, drug discovery, and translational medicine, with more Life Sciences models promised to follow. The name honors Rosalind Franklin, the British chemist whose X-ray crystallography work helped crack DNA's double helix structure, only to get systematically written out of the history books during her lifetime. Bringing a drug from target discovery to FDA approval typically runs 10 to 15 years, experts say, with most of that time spent not on flashes of inspiration but on grinding through reams of literature, mining databases, crafting reagents, and untangling results that could go either way. GPT-Rosalind is positioned as a tool to help researchers "explore more possibilities, surface connections that might otherwise be missed, and arrive at better hypotheses sooner," per the company announcement.

The numbers at least give the hype some legroom. On BixBench—a benchmark designed around real-world bioinformatics challenges—GPT-Rosalind posted a 0.751 pass rate, snagging the top spot among models with published results. On LABBench2, it bested its predecessor GPT-5.4 on six of eleven tasks. OpenAI also announced that Dyno Therapeutics will stress-test the model against unpublished RNA sequences to rule out old-fashioned memorization. In sequence prediction tasks, the company's best-of-ten submissions cleared the 95th percentile of human experts, hovering around the 84th percentile on generation. Still, OpenAI's life sciences research lead Joy Jiao kept things grounded: "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," she told the LA Times during a press briefing.

The infrastructure around the model might matter just as much as the model itself. OpenAI is rolling out a free Life Sciences research plugin for Codex that hooks into over 50 scientific databases and tools—protein structure lookups, sequence searches, literature reviews, genomics pipelines, the whole buffet. Enterprise customers with GPT-Rosalind access get the reasoning layer on top, while everyone else gets the plugin paired with standard models. For the launch, OpenAI assembled a pharma dream team: Amgen, Moderna, and Thermo Fisher Scientific are all in. There's also a research partnership with Los Alamos National Laboratory focused on AI-guided protein and catalyst design. "The life sciences field demands precision at every step. The questions are highly complex, the data are highly unique, and the stakes are incredibly high," said Sean Bruich, Amgen's Senior VP of AI and Data, in the official announcement.

Access to Rosalind, unsurprisingly, comes with strings attached. U.S. enterprise only, gated behind a qualification and safety review—an arrangement that mirrors broader hand-wringing in the scientific community. An international coalition of over 100 scientists has already pushed for stricter controls on biological data used to train AI, flagging pathogen design risks. During the research preview phase, usage won't eat into existing API credits. This isn't OpenAI's first rodeo into science workflows; the company launched its Prism scientific writing workspace back in January. Zero fully AI-discovered drugs have cleared phase 3 trials so far. But if GPT-Rosalind helps a researcher cook up a better experiment six months ahead of schedule across thousands of labs, the compounding returns on what gets discovered and when could get genuinely interesting.