The Pulse
OpenAI Moves GPT-Rosalind Into Global Life Sciences Release
OpenAI has moved GPT-Rosalind out of research preview and made the life sciences model available globally to eligible organizations through its trusted-access program. The model supports biology, genomics, drug discovery, protein analysis a
AI.info Team ·
OpenAI has moved GPT-Rosalind from research preview into a global release for eligible organizations, giving approved life sciences teams access to the company’s specialized model through a trusted-access program. The update, posted September 11, also sets October 5 as the date when published pricing will take effect.
GPT-Rosalind is designed for research across biology, drug discovery and translational medicine. OpenAI says the model is available through ChatGPT, Codex and the API, while access remains subject to organizational review, security controls and governance requirements. The company first introduced GPT-Rosalind on April 16 as the opening model in a new life sciences series.
The release marks a shift from a limited preview aimed mainly at selected research partners to a product offering that OpenAI can make available to qualified organizations worldwide. It does not turn GPT-Rosalind into a general consumer chatbot. Access continues to be restricted to organizations conducting legitimate scientific research with public-benefit goals and safeguards against misuse.
GPT-Rosalind Leaves Research Preview
OpenAI’s September 11 update says GPT-Rosalind is “coming out of research preview” and is now available globally to eligible organizations through the company’s trusted-access program. Eligible organizations will continue to receive access to the latest Rosalind models as OpenAI releases them, according to the company’s announcement.
The change broadens the geographic scope of the initial launch. OpenAI originally described the model as launching first to qualified Enterprise customers in the United States, with controls covering eligibility, access management and organizational governance. The latest announcement expands the program beyond that initial U.S. starting point while retaining the review process.
OpenAI has also introduced a managed workspace for qualified organizations that do not have an Enterprise account. The company describes that workspace as an option for approved users who need access to GPT-Rosalind without adopting a full Enterprise arrangement. The model’s API remains subject to eligibility review and organizational approval.
Release status: GPT-Rosalind is globally available to eligible organizations through trusted access. Published pricing takes effect on October 5, 2026, while access remains controlled rather than open to the general public.
From Literature Review to Experimental Planning
GPT-Rosalind is built for research tasks that require reasoning across scientific evidence, biological data, specialist databases and experimental results. OpenAI identifies chemistry, protein engineering and genomics as central areas for the model, with workflows spanning literature synthesis, sequence-to-function interpretation, experimental design and data analysis.
The model is intended to help researchers compare biological targets, interpret genes and pathways, examine protein structures and variants, assess molecular candidates and plan follow-up experiments. OpenAI’s Rosalind product page describes a workflow in which scientists can bring together published papers, trusted biology data and internal results to identify areas of agreement, expose missing evidence and determine what to test next.
Another workflow focuses on sequencing analysis. Researchers can provide sequencing files and sample information, review an analysis plan, monitor quality-control results and inspect saved outputs before deciding on the next biological question. OpenAI says the Rosalind Workbench can connect these activities to scientific tools and data sources in a shared environment that teams can review and reuse.
The practical emphasis separates GPT-Rosalind from a model marketed only as a scientific question-answering system. OpenAI is positioning it as a system that can coordinate several steps in a research process, provided users approve the plan, inspect the outputs and keep scientific judgment in the loop.
OpenAI Reports Gains on Biology Benchmarks
OpenAI reports that GPT-Rosalind outperforms GPT-5.4 on six of 11 tasks in LABBench2, a benchmark covering literature retrieval, database access, sequence manipulation and protocol design. The company identifies CloningQA as the largest improvement in that evaluation; the task requires end-to-end design of DNA and enzyme reagents for molecular cloning protocols.
OpenAI also evaluated the model with Dyno Therapeutics, which works on AI-designed gene therapies. The test used unpublished RNA sequences that were not part of the model’s training data and compared GPT-Rosalind with 57 historical scores from human experts in AI and biology. In the Codex application, OpenAI says the best of ten model submissions ranked above the 95th percentile of human experts on RNA sequence-to-function prediction and around the 84th percentile on sequence generation.
Those figures come from company-described evaluations rather than an independent clinical or laboratory validation study. They indicate performance on selected research tasks, not the ability to discover a safe drug, establish a biological mechanism or replace experimental confirmation. OpenAI’s own product description frames the model as support for evidence synthesis, hypothesis generation and planning rather than as an autonomous scientific authority.
The separate Rosalind product page lists reported performance-per-token increases of 53.7% on Genebench, 18.0% on Medchem Bench, 19.6% on Labworkbench and 4.42% on LifeSci Bench. OpenAI does not present those figures as clinical outcomes. They are model evaluation results tied to specific life sciences tasks.
The Codex Plugins Add a Research Execution Layer
OpenAI is releasing a Life Sciences Research plugin for Codex through GitHub. The package contains modular skills for human genetics, functional genomics, protein structure, biochemistry, clinical evidence and public-study discovery, and connects models to more than 50 public multi-omics databases, literature sources and biology tools.
The plugin is available for broader use with OpenAI’s mainline models, while eligible Enterprise users can combine it with GPT-Rosalind for deeper biological reasoning. OpenAI describes the software as an orchestration layer for broad, ambiguous and multi-step questions rather than as a single database or standalone scientific application.
A separate Life Sciences NGS Analysis plugin is designed to create auditable outputs from sequencing workflows. OpenAI says the tool can validate inputs, return quality-control information, produce Salmon expression matrices and preserve provenance and caveats in a run record that researchers can inspect and reuse in Codex.
That audit trail matters because biological research often depends on the details of how an analysis was performed, not only on the final answer. A model that produces a conclusion without showing the source data, software steps and limitations can create additional work for scientists. OpenAI’s stated design places reviewable outputs and explicit caveats inside the workflow.
Novo Nordisk Joins OpenAI’s Global Expansion
OpenAI says it is working with pharmaceutical, biotechnology and research organizations including Amgen, Novo Nordisk, Thermo Fisher Scientific, Moderna, the Allen Institute, Oracle Health and Life Sciences, NVIDIA, Benchling and the UCSF School of Pharmacy.
The September update highlights Novo Nordisk’s participation in the expanded program. OpenAI says the Danish drugmaker is using GPT-Rosalind to help researchers analyze complex datasets, identify useful patterns and test hypotheses more quickly.
“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,” said Mishal Patel, group vice president for AI and digital innovation in research and development at Novo Nordisk.
Patel said Novo Nordisk is exploring GPT-Rosalind as part of a partnership with OpenAI focused on more rigorous and practical approaches to drug discovery. OpenAI has not announced a drug candidate, regulatory submission or clinical result produced by the partnership.
Access Controls Reflect Biology’s Dual-Use Risk
OpenAI’s deployment plan treats biological capability as a safety issue as well as a product feature. The company says participating organizations must conduct legitimate research with clear public benefit, maintain governance and misuse-prevention controls, and restrict access to approved users in secure environments.
The GPT-Rosalind-5.5 system card says the model is incrementally trained from GPT-5.5 with additional data related to beneficial life sciences capabilities. Unlike GPT-5.5, it is trained not to refuse sophisticated biology queries, with trusted access and responsible deployment serving as the primary safeguard.
OpenAI classifies GPT-Rosalind-5.5 as having high capability in its biological and chemical preparedness category, while saying it falls below the threshold for critical capability. The company also says its Safety Advisory Group approved the safeguards plan for the release after assessing the risk of severe harm.
Those statements explain why the model is not offered as an unrestricted public endpoint. The same scientific knowledge that can help with medicinal chemistry, genomics and laboratory troubleshooting can also create risks if placed in the hands of users seeking harmful biological assistance. OpenAI says GPT-Rosalind is trained to refuse malicious requests that would meaningfully enable biological weaponization.
Rosalind’s broader product program includes a biodefense initiative for trusted developers and public-health teams. OpenAI lists early detection, preparedness, diagnostics, response and medical countermeasures as focus areas. Access for those applications also depends on review and public-benefit boundaries.
Pricing Begins October 5, but the Research Questions Remain
OpenAI has not included published pricing details in its September 11 announcement, but it says published pricing will take effect on October 5, 2026. The company’s move out of preview therefore changes commercial availability before it fully discloses the public cost structure in the announcement itself.
The central test will be whether GPT-Rosalind improves the quality and speed of work that scientists can verify. Benchmark gains and partner statements provide an early signal, but they do not show whether the model can raise the success rate of drug programs, reduce failed experiments or produce findings that survive independent replication.
OpenAI’s own material leaves several practical questions for research organizations. How will teams measure results across proprietary datasets? Which workflows will remain dependent on human review? How will access controls adapt as the model gains stronger biological reasoning, and what evidence will show that the system’s use produces better decisions rather than faster production of plausible but incorrect hypotheses?
For now, the concrete change is narrower: as of September 11, GPT-Rosalind is no longer limited to research preview. Eligible organizations around the world can request access, and published pricing is scheduled to begin on October 5.