OpenAI drops GPT-Rosalind to speed up drug discovery

Developing a new drug in the U.S. takes between 10 and 15 years, and much of that time is spent grinding through literature, messy biological databases, and experimental protocols. OpenAI wants to shorten that first stretch.

The AI lab just launched GPT-Rosalind, a frontier reasoning model designed specifically for life sciences research, drug discovery, and translational medicine. Named after Rosalind Franklin, the scientist whose work was vital to uncovering the double-helix structure of DNA, the specialized model tackles chemistry, protein engineering, and genomics workflows.

Access is rolling out through a trusted-access program. Initially limited to qualified U.S. enterprise customers, OpenAI later expanded global access to eligible organizations and set official pricing to take effect on October 5, 2026.

Beating human experts on RNA workflows

General frontier models are great at summarizing papers or writing Python scripts, but they routinely stumble when tasked with complex biochemical reasoning. GPT-Rosalind is built to reason over molecular mechanisms, protein mutations, and disease pathways.

In public benchmarks, the model posted top scores on BixBench, an evaluation focused on real-world bioinformatics. On LABBench2, which covers end-to-end biological research tasks, GPT-Rosalind beat GPT-5.4 on 6 out of 11 challenges, showing its biggest gains on CloningQA by designing complete molecular cloning protocols with enzymes and DNA reagents.

OpenAI also teamed up with gene-therapy company Dyno Therapeutics to test the model on uncontaminated, unpublished RNA data. Running directly inside the Codex app, the best submission out of ten attempts beat 95% of human experts on sequence-to-function predictions, and ranked in the 84th percentile on sequence generation tasks.

50 databases and deep pharma partnerships

To plug into real lab work, OpenAI released an open-source Life Sciences research plugin for Codex on GitHub. It connects AI models to more than 50 public multi-omics databases, biological tools, and scientific literature sources.

While GPT-Rosalind itself is gated behind strict enterprise safety reviews, any developer or scientist can use the plugin with OpenAI mainline models for tasks like sequence search and protein structure lookups.

OpenAI is rolling this out with major life sciences heavyweights, including Amgen, Moderna, Thermo Fisher Scientific, and the Allen Institute. It is also collaborating with Los Alamos National Laboratory on AI-guided catalyst and protein design, while consulting firms McKinsey, BCG, and Bain help integrate the model into pharma pipelines.

Why it matters

Life sciences research is perhaps the most lucrative proving ground for AI reasoning, and OpenAI is taking a calculated angle. Instead of going head-to-head with DeepMind on raw molecular folding like AlphaFold, OpenAI is positioning GPT-Rosalind as an intelligent orchestration layer that can hypothesize, query 50 databases, and write wet-lab protocols in one go. For pharma researchers, this turns multi-day planning and manual database aggregation into minutes of reasoning, compounding early discovery gains.

We think OpenAI's strategy here is clever: partnering with top consulting firms and premier biotechs builds an immediate enterprise moat where budgets are massive. However, the stakes in life sciences are completely unforgiving compared to software coding. A hallucinated primer design or a flawed enzymatic reaction can cost labs tens of thousands of dollars and months of lost experimental time. Restricting access through trusted channels is prudent for biosecurity, but the ultimate hurdle will be convincing wet-lab scientists that a reasoning model's biological outputs are rigorously dependable.