en.Wedoany.com Reported - On April 16 local time, OpenAI announced the launch of GPT-Rosalind, a cutting-edge reasoning model specifically built to support biology, drug discovery, and translational medicine research. Named after British chemist Rosalind Franklin, it is the first model in OpenAI's life sciences series. GPT-Rosalind is available as a research preview to eligible customers via ChatGPT, Codex, and APIs, and does not consume existing usage quotas during the preview period.
GPT-Rosalind is positioned as a specialized reasoning engine for the life sciences field. According to Yunyun Wang, OpenAI's Head of Life Sciences Products, the model aims to address two major core obstacles faced by biology researchers: first, the massive amount of data accumulated from decades of genome sequencing and protein biochemistry exceeds individual processing capabilities; second, the highly specialized subfields within biology create knowledge barriers. To this end, OpenAI trained the model on 50 common biological workflows based on a general large language model and integrated mainstream public databases. This enables the model to connect genotype and phenotype through known pathways and regulatory mechanisms, infer protein structure or functional properties, and screen for potential drug targets. Joy Jiao, OpenAI's Head of Life Sciences Research, emphasized that the model's goal is not to replace scientists, but to help researchers accelerate the most complex and time-consuming parts of the scientific process.
Performance evaluations show that GPT-Rosalind performs exceptionally well in multiple benchmark tests. According to VentureBeat, GPT-Rosalind achieved the highest score among publicly disclosed models on the bioinformatics benchmark BixBench. In the LABBench2 benchmark covering 11 research tasks including literature retrieval, sequence manipulation, and experimental protocol design, GPT-Rosalind outperformed GPT-5.4 in 6 tasks, with the most significant improvement seen in molecular cloning experimental design tasks. In a joint evaluation with AI gene therapy company Dyno Therapeutics, GPT-Rosalind used unpublished RNA sequences not contaminated by training data for sequence function prediction. The best result from ten submissions ranked above the 95th percentile of historical human expert performance, while sequence generation ranked around the 84th percentile.
Access is strictly controlled. OpenAI has deployed GPT-Rosalind under a trusted access framework, currently limited to entities based in the United States for application and use. Yunyun Wang stated that restricting usage permissions aims to maximize research value while minimizing the risk of misuse. The system has a built-in high-precision flagging mechanism that automatically issues warnings if users touch upon specific indicators or thresholds related to biological weapons. The first batch of partner clients includes Amgen, Moderna, the Allen Institute, and Thermo Fisher Scientific. OpenAI is also collaborating with Los Alamos National Laboratory to jointly explore AI-guided protein and catalyst design.
Concurrently, OpenAI has open-sourced the Codex Life Sciences Research plugin on GitHub, which integrates over 50 public multi-omics databases, literature sources, and bioinformatics tools, covering areas such as human genetics, functional genomics, protein structure, and biochemistry. This plugin is free for all users, is not limited to GPT-Rosalind, and can also be used with general-purpose models. Explaining the choice of life sciences as the first vertical industry, OpenAI provided a quantitative rationale: on average, it takes 10 to 15 years for a new drug to progress from target discovery to market approval, and only one in ten drugs entering clinical trials is ultimately approved. Acceleration by AI in the early discovery phase can have a compound effect downstream, improving the success rate of the entire pipeline.
Competition among tech giants in the life sciences AI arena continues to intensify. Two Google DeepMind scientists won the 2024 Nobel Prize in Chemistry for the protein structure prediction system AlphaFold. DeepMind's Isomorphic Labs launched the computational drug design system IsoDDE in February 2026. OpenAI, Anthropic, and Google have significantly increased their investments in AI for healthcare and scientific applications in recent years. However, real-world challenges remain: according to information from OpenAI's launch event, no drug fully discovered or designed by AI has yet passed Phase III clinical trials, with only a few entering early clinical stages. An international team of over a hundred scientists has already called for stricter controls on sensitive biological data used to train AI systems to prevent the technology from being misused to create dangerous pathogens.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
