Credit: Shutterstock
OpenAI recently introduced GPT-Rosalind, a powerful reasoning model built from the ground up for biology, drug discovery, and translational medicine.
Let me explain the need for it first.
People from the life sciences field will second me that getting a drug from discovery to regulatory approval takes an average of 10 to 15 years. The majority of this time goes into parsing literature, querying databases, reconciling ambiguous experimental results, and maintaining a coherent biological hypothesis across months of fragmented workflows. This is a lot of time during which the drug could have helped many patients.
GPT-Rosalind will remove this grind and save you time.
OpenAI didn’t just fine-tune an existing model and give it a new name. GPT-Rosalind is an all-new release in a dedicated Life Sciences model series and the first time OpenAI has built a domain-specific reasoning model for biology. Interestingly, it’s also the first time they’ve named a model after a real historical figure, “Rosalind Franklin.”
Who Was Rosalind Franklin?
Rosalind Franklin was a British chemist whose X-ray crystallography in the 1950s was foundational in determining the double-helix structure of DNA. She wasn’t credited for it during her lifetime. Giving credit where it’s due, OpenAI named this model after her, showing how the company respects those who served the human race and remained unrecognized. It’s a good choice.
How GPT-Rosalind Works
At its core, GPT-Rosalind is a reasoning model optimized for multi-step scientific workflows. It can synthesize evidence from literature, generate and evaluate biological hypotheses, design experiments, query genomic and proteomic databases, and interpret complex experimental outputs, all within a single workflow.
It’s built for tasks that sit at the intersection of biochemistry, protein engineering, genomics, and translational medicine. The kinds of tasks where a researcher might currently spend days jumping between PubMed, UniProt, BLAST, and a dozen other tools. GPT-Rosalind saves the researchers a lot of time.
What the Benchmarks Say
On BixBench, a benchmark based on real-world bioinformatics and data analysis tasks, GPT-Rosalind scored 0.751, the highest published score on that benchmark to date.
On LABBench2, which covers tasks such as literature retrieval, database access, sequence manipulation, and protocol design, GPT-Rosalind outperformed GPT-5.4 on 6 of 11 tasks.
Then there’s the Dyno Therapeutics evaluation. Dyno, a gene therapy company, tested GPT-Rosalind on RNA sequence-to-function prediction and generation using unpublished sequences specifically to rule out memorization. The model’s best-of-ten submissions were compared against 57 historical scores from human experts in the AI-biology field. On the prediction task, the model ranked above the 95th percentile, and on the sequence generation task, around the 84th percentile.
Those are strong numbers. But Joy Jiao, OpenAI’s life sciences research lead, was careful at the press briefing not to overread them. The model isn’t designed to develop new drugs autonomously. The goal is speed and helping researchers move through the most complex and time-intensive parts of the scientific process faster.
This makes sense because no AI-discovered drug has cleared Phase 3 clinical trials. But compressing early-stage research cycles across thousands of labs simultaneously will save a lot of time, and that’s where the real long-term value lies.
Who Has Access and Why It’s Restricted
GPT-Rosalind is not publicly available. Right now, it’s limited to US enterprise customers who go through a qualification and safety review. That’s a deliberate decision, and an understandable one.
More than 100 scientists signed an open letter calling for tighter controls on biological data used to train AI models, citing the risk of dual-use misuse, specifically the design of pathogens. OpenAI’s response is a tiered access structure intended only for beneficial purposes, under strong governance and with enterprise-grade access management.
Organizations need to conduct legitimate scientific research with clear public benefit, maintain compliance and misuse-prevention controls, and restrict access to approved users within governed environments.
During the research preview, use of the model doesn’t consume credits or tokens.
Alongside the main model, OpenAI is releasing a free Life Sciences research plugin for Codex, now available on GitHub. It’s not the same as GPT-Rosalind, but it’s a meaningful tool for a researcher who currently has to string these tools together manually.
The plugin connects to more than 50 scientific databases and tools, covering protein structure, functional genomics, sequence search, literature review, and public dataset discovery. It works as an orchestration layer for common research workflows: protein structure lookup, BLAST-style sequence search, clinical evidence retrieval, and more.
The Bigger Picture
GPT-Rosalind is OpenAI’s first move into domain-specific AI for science. It won’t replace a medicinal chemist or a structural biologist. It’s not supposed to. What it’s supposed to do is make the early-stage discovery process faster, less fragmented, and more scalable so that researchers can run more hypotheses, surface more connections, and arrive at better experiments sooner.
