OpenAI Unveils GPT-Rosalind for Life Sciences and Drug Discovery

OpenAI has officially introduced GPT-Rosalind, its first specialized AI model designed specifically for the life sciences. Named in honor of pioneering chemist Rosalind Franklin, the model is engineered to provide advanced reasoning capabilities in fields such as biochemistry, protein engineering, and genomics. By moving beyond the broad training of general-purpose language models, GPT-Rosalind is fine-tuned to assist researchers in navigating the complex, time-intensive analytical demands of modern biological discovery.

Accelerating Scientific Workflows

Drug discovery is a notoriously lengthy process, often requiring 10 to 15 years to progress from initial target discovery to regulatory approval. GPT-Rosalind aims to compress these timelines by supporting multi-step research tasks, including evidence synthesis, hypothesis generation, and experimental planning. The model allows researchers to query specialized databases, parse scientific literature, and interact with computational tools within a single interface, effectively streamlining the early stages of the scientific process.
To further support these workflows, OpenAI is launching a Life Sciences research plugin for Codex. This tool provides programmatic access to over 50 scientific data sources and computational pipelines, enabling researchers to integrate the model directly into their existing technical environments.

Benchmark Performance and Real-World Validation

OpenAI has validated the model’s capabilities through both standardized benchmarks and practical research applications. On BixBench, a benchmark focused on bioinformatics and data analysis, GPT-Rosalind achieved a 0.751 pass rate. Furthermore, the model outperformed GPT-5.4 on six out of eleven tasks in LABBench2, showing significant proficiency in CloningQA, which involves the end-to-end design of reagents for molecular cloning protocols.
In a real-world evaluation conducted in partnership with Dyno Therapeutics, the model was tested on RNA sequence-to-function prediction using unpublished data. The results were notable, with the model’s best-of-ten submissions ranking above the 95th percentile of human experts on prediction tasks and reaching the 84th percentile for sequence generation.

Controlled Access and Strategic Partnerships

Access to GPT-Rosalind is currently gated through a trusted-access program for qualified enterprise organizations in the United States. To ensure safety and security, OpenAI has implemented technical safeguards to monitor for potentially dangerous activity and has established strict governance controls. The model is available via ChatGPT, Codex, and the OpenAI API.
OpenAI is already collaborating with several prominent organizations, including Amgen, Moderna, the Allen Institute, and Thermo Fisher Scientific, to integrate the model into their research workflows. Additionally, the company is partnering with the Los Alamos National Laboratory to explore AI-guided design of proteins and catalysts. This launch represents a shift toward domain-specific AI, prioritizing models optimized for high-dimensional scientific data and complex search spaces.