OpenAI Launches GPT-Rosalind for Life Sciences
OpenAI unveiled GPT-Rosalind, a new reasoning model created for life sciences research. This tool supports researchers in areas such as combining evidence, creating hypotheses, designing experiments, and analyzing data. The model aims to speed up the process from initial ideas to practical tests.
Background and Naming
The model takes its name from chemist Rosalind Franklin. Franklin's X-ray diffraction images played a key role in revealing DNA's double helix structure in the 1950s, alongside work by James Watson and Francis Crick. GPT-Rosalind focuses on challenges in biosciences, drug development, and moving research into medical applications. Researchers can use it to gather evidence from studies, form new ideas, outline experiments, and manage complex tasks with multiple steps.
OpenAI tuned GPT-Rosalind for scientific processes. It reasons more precisely about molecules, proteins, genes, cell signaling paths, and biology linked to diseases. The model also integrates scientific databases and tools better in extended workflows. Tasks it handles cover searching literature, understanding links between genetic sequences and their functions, planning experiments, and processing data.
Strong Benchmark Results
OpenAI's tests show GPT-Rosalind surpassing GPT-5, GPT-5.2, and GPT-5.4. It leads in chemistry, biochemistry and protein knowledge, phylogenetics, experiment design and analysis, and tool use. Gains were largest in experiment design and chemistry.
On the public BixBench benchmark for bioinformatics and data analysis, GPT-Rosalind achieved a Pass@1 score of 0.751. That result tops GPT-5.4 at 0.732, Grok 4.2 at 0.698, GPT-5 at 0.728, and Gemini 3.1 Pro at 0.550. OpenAI reports further details placing it ahead of GPT-5.4 at 0.732, Grok 4.2 at 0.728, GPT-5.2 at 0.698, and GPT-5 at 0.611.
For LABBench2, which tests literature search, database use, sequence handling, and protocol creation, GPT-Rosalind outperformed GPT-5.4 on 6 of 11 tasks. The most notable improvement appeared in CloningQA, a task for designing full DNA and enzyme setups in molecular cloning protocols.
Free Research Plugin
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
OpenAI also launched a free plugin for Codex on GitHub. This life sciences tool offers building blocks for typical research steps. It links models to over 50 public databases and resources in multi-omics, literature, and biology, including human genetics, functional genomics, protein structures, biochemistry, clinical data, and study finding. Examples include AlphaFold for protein folding predictions, Bgee for gene expression, and BindingDB for molecular interactions.
The plugin acts as a coordinator for wide-ranging, unclear, multi-part queries. Enterprise users pair it with GPT-Rosalind. Others connect it to regular OpenAI models.
Limited Access and Future Outlook
GPT-Rosalind starts as a research preview in ChatGPT, Codex, and the API. Only qualified US enterprise customers qualify through the Trusted Access Program. Preview use does not count against current credits or tokens. OpenAI plans to announce pricing and wider access soon.
Organizations need to meet three criteria for entry. They must conduct real scientific work with clear benefits to the public. They require strong governance, compliance, and measures against misuse. Access must restrict to cleared users in protected, controlled settings.
GPT-Rosalind marks the start of a model line for life sciences. OpenAI intends to grow its abilities in biochemical reasoning for research with heavy tool use and long sequences.
Early users and partners include Amgen, a biotech leader in protein therapies; Novo Nordisk, focused on diabetes treatments; Moderna, known for mRNA vaccines; Thermo Fisher Scientific, a lab equipment giant; Oracle Health and Life Sciences; NVIDIA, with AI hardware; the Allen Institute for brain science; Benchling, a biotech software firm; and the UCSF School of Pharmacy.