OpenAI has updated its GPT-Rosalind model for life sciences research and made it available in research preview to eligible organisations worldwide.
The revised system is part of OpenAI's GPT-Rosalind series for scientific work in areas including drug discovery, genomics and laboratory analysis. The update combines features from GPT-5.5 with stronger performance in medicinal chemistry, quantitative biology and wet lab troubleshooting.
OpenAI also disclosed a broader rollout through what it described as a trusted-access deployment structure. Qualified organisations without an Enterprise account can use an OpenAI-managed workspace.
Benchmark results
To measure the model's performance, OpenAI introduced several benchmarks covering different parts of life sciences research. It built LifeSciBench as an externally judged benchmark spanning evidence handling, analysis, design and optimisation, scientific reasoning, validation and operations, and translation and communication.
OpenAI said GPT-Rosalind led across tasks identified by academic and industry experts, including the extraction and reconciliation of scientific evidence from papers, figures, tables and experimental records.
In medicinal chemistry, OpenAI created MedChemBench to test capabilities such as chemical structure understanding, structure-activity relationships, potency and toxicity prediction, lead-optimisation decisions and retrosynthesis. GPT-Rosalind scored 27.5% on MedChemBench, compared with 25.1% for GPT-5.5, while using 7.2% fewer tokens.
Results in genomics and quantitative biology were reported through a separate evaluation called GeneBench. OpenAI said GPT-Rosalind reached 21.6% accuracy, compared with 20.4% for GPT-5.5, and used 31% fewer tokens on long-horizon analysis tasks covering fields such as functional genomics, spatial transcriptomics, proteomics, epigenomics and applied genetics.
OpenAI also introduced LabWorkBench, a benchmark intended to test whether the model can assist with practical laboratory work, including troubleshooting and optimisation of real wet lab protocols. GPT-Rosalind scored 63.2%, versus 55.8% for GPT-5.5, while using 5.3% fewer tokens.
Workflow tools
Alongside the model update, OpenAI built two plugins, Life Sciences Research and Life Sciences NGS Analysis, to let researchers carry out repeatable scientific workflows within Codex. The tools are intended to combine evidence retrieval, biological interpretation and bioinformatics execution in a single workspace.
According to OpenAI, all users can access both plugins through Codex. Qualified GPT-Rosalind users can also use the model with those plugins.
OpenAI has added interactive viewers for biological file types including sequence, alignment and structure data. The viewers are intended to let scientists inspect evidence during a workflow and ask follow-up questions with the active file in context.
The company gave examples of tasks handled through the NGS Analysis plugin, including single-cell RNA sequencing quality control and annotation, and bulk RNA sequencing quality control from FASTQ files. It said the plugin can route requests, validate inputs, preserve provenance and return auditable outputs such as MultiQC reports and count matrices.
Access expansion
OpenAI said access to GPT-Rosalind is being expanded to eligible organisations conducting scientific research with public benefit, governance and safety oversight, and controlled access arrangements. As part of that expansion, it named Novo Nordisk as one of the organisations using the system.
According to OpenAI, the Danish drugmaker is using GPT-Rosalind in medical research, with a focus on analysing complex datasets, identifying patterns and testing hypotheses across literature, genomics, transcriptomics, sequence, structure and experimental results.
"Life sciences research is complex, data-rich, and interdisciplinary. To deliver meaningful value for researchers, advanced AI models must be grounded in trusted scientific data, connected to validated tools, and integrated into the real-world workflows researchers use every day. We're pleased with our partnership with OpenAI and the opportunity to explore how GPT‐Rosalind can support more rigorous, practical approaches to drug discovery," said Mishal Patel, Group Vice President, AI & Digital Innovation, R&D - Novo Nordisk.
OpenAI said the model update forms part of its broader work in life sciences, including applications in translational medicine, public health, preparedness and biodefence. The GPT-Rosalind programme sits within a controlled deployment approach aimed at qualified research organisations.