Named after the crystallographer who helped reveal the construction of DNA, GPT-Rosalind is OpenAI’s first domain-specific mannequin collection, fine-tuned for biochemistry, genomics, and protein engineering. Entry is restricted to a trusted-access programme for vetted enterprise clients together with Amgen, Moderna, and Thermo Fisher Scientific.
OpenAI has launched GPT-Rosalind, a frontier reasoning mannequin constructed particularly for all times sciences analysis, the corporate introduced on Thursday.
The mannequin is designed to assist proof synthesis, speculation era, experimental planning, and multi-step scientific workflows throughout biochemistry, genomics, and protein engineering, representing OpenAI’s first purpose-built domain-specific mannequin collection.
It’s accessible as a analysis preview in ChatGPT, Codex, and the OpenAI API, however entry is restricted to a trusted-access programme for certified enterprise clients in the US.
The 💜 of EU tech
The newest rumblings from the EU tech scene, a narrative from our smart ol’ founder Boris, and a few questionable AI artwork. It is free, each week, in your inbox. Enroll now!
The mannequin is called after Rosalind Franklin, the British chemist and X-ray crystallographer whose diffraction imaging of DNA was instrumental in revealing the double helix construction, and whose contribution was notably absent from the 1962 Nobel Prize awarded to Watson, Crick, and Wilkins.
The naming is a pointed act of recognition: Franklin’s work is now extensively thought to be foundational to trendy molecular biology, and he or she stays a touchstone in discussions in regards to the erasure of ladies from scientific historical past.
OpenAI is framing GPT-Rosalind as a software to compress the timeline from scientific concept to medical proof. The corporate estimates it presently takes roughly 10 to fifteen years to maneuver a drug from goal discovery to regulatory approval in the US.
GPT-Rosalind is positioned to assist on the early phases: it could question specialised databases, parse scientific literature, work together with computational instruments, and counsel new experimental pathways inside a single interface.
Alongside the mannequin itself, OpenAI can be introducing a Life Sciences analysis plugin for Codex that connects fashions to greater than 50 scientific instruments and information sources, giving researchers programmatic entry to organic databases and computational pipelines.
Launch companions embody Amgen, Moderna, Thermo Fisher Scientific, and the Allen Institute. OpenAI can be working with Los Alamos Nationwide Laboratory on AI-guided protein and catalyst design.
Benchmark efficiency, as reported by OpenAI, exhibits GPT-Rosalind attaining a 0.751 move price on BixBench, a bioinformatics benchmark developed by Edison Scientific that evaluates fashions on real-world computational biology duties.
On LABBench2, a broader analysis job benchmark, the model outperformed GPT-5.4 on six of eleven tasks, with its most vital benefit on CloningQA, a job requiring the end-to-end design of reagents for molecular cloning protocols.
Essentially the most placing efficiency sign got here from a third-party analysis performed with Dyno Therapeutics, a gene remedy firm targeted on designing AAV capsid proteins.
Utilizing unpublished, beforehand unseen RNA sequences to protect towards benchmark contamination, GPT-Rosalind was examined on sequence-to-function prediction and sequence era duties.
The very best-of-ten mannequin submissions ranked above the ninety fifth percentile of human consultants on the prediction job and across the 84th percentile on sequence era, in accordance with OpenAI and confirmed by a number of shops protecting the launch.
The launch carries important dual-use caveats that OpenAI has addressed by way of its entry mannequin. Researchers have warned that AI fashions skilled on organic information might be misused to assist design harmful pathogens.
OpenAI’s determination to limit entry solely to a vetted trusted-access programme, with organisations required to reveal they’re working in the direction of bettering human well being outcomes and sustaining robust safety and governance controls, is a direct response to that threat. In the course of the analysis preview part, utilization won’t eat present API credit.
