OpenAI Brings GPT-Rosalind to U.S. University Biology Labs for Free
OpenAI is offering eligible U.S. university labs up to 12 months of sponsored access to its life-sciences reasoning model, GPT-Rosalind.
- OpenAI launched the Rosalind Academic Plan, offering sponsored GPT-Rosalind access to eligible U.S. university labs.
- Qualifying labs get up to 12 months of free access, with GPT-Rosalind requiring additional trusted-access approval.
- Research data shared through the plan is not used to train OpenAI models by default.
- GPT-Rosalind beats GPT-5.4 on 6 of 11 LABBench2 tasks, led by DNA cloning design.
- The model is already used by Amgen, Moderna, Novo Nordisk, Thermo Fisher, and the Allen Institute.
- Program targets early-stage biology research in drug design, protein engineering, and neurodegeneration.
OpenAI brings GPT-Rosalind to U.S. university labs
OpenAI has launched the Rosalind Academic Plan, which gives eligible labs at U.S. universities sponsored access to GPT-Rosalind and other OpenAI tools for up to 12 months. The program lowers the cost of testing a specialized biology model, although access requires a separate trusted-access review.
The plan extends a model previously available through enterprise agreements to academic researchers working on basic biology. Participating labs can use it for literature analysis, database retrieval, sequence work and experimental planning, subject to OpenAI’s safeguards for sensitive biological research.
Built for biology’s long workflows
OpenAI describes GPT-Rosalind as its first domain-specific model. The company developed it for multi-step work in drug discovery, genomics, protein research and molecular biology, where a useful answer may require combining papers, structured databases, sequence data and protocol constraints.
The model is named after Rosalind Franklin, the chemist and X-ray crystallographer whose research helped reveal the molecular structure of DNA and viruses. OpenAI’s model overview positions GPT-Rosalind as a reasoning system for scientific workflows rather than a general-purpose assistant with biology knowledge.
What the benchmark numbers show
OpenAI reports results from LABBench2, a benchmark covering literature retrieval, database access, sequence manipulation and protocol design. The company also cites an external evaluation from Dyno Therapeutics.
| Evaluation | Reported result | What it measures |
|---|---|---|
| LABBench2 | GPT-Rosalind exceeded GPT-5.4 on 6 of 11 tasks | Scientific retrieval, data access, sequence work and protocol design |
| CloningQA | OpenAI reports its largest LABBench2 improvement on this task | End-to-end design of DNA constructs, enzymes and molecular-cloning protocols |
| Dyno Therapeutics evaluation | Best-of-10 outputs ranked above the 95th percentile of expert biologists | Prediction of RNA function |
Best-of-10 scoring reports the strongest result from 10 attempts, so it should not be read as expected single-response performance. Benchmark results also leave open how reliably the model performs across unfamiliar organisms, incomplete datasets and live laboratory constraints. Experimental validation and expert review remain necessary.
One year, with a second gate
Eligible academic groups receive several benefits and restrictions under the plan:
- Sponsored access to OpenAI tools for up to 12 months
- GPT-Rosalind access after a separate trusted-access approval
- A default policy that excludes submitted research data from model training
- Eligibility limited to labs at U.S. universities
GPT-Rosalind has also moved beyond research preview and is available globally to approved organizations through the trusted-access program. Those customers receive updated Rosalind models as OpenAI releases them, with commercial pricing scheduled to apply in October. The academic program covers costs temporarily for a narrower group.
Where labs may save time
The model’s clearest use cases involve research tasks that require repeated searches, cross-checks and constraint tracking. OpenAI highlights workflows such as:
- Collecting physicochemical properties for candidate drugs from PubChem, NCBI and related databases
- Designing molecular-cloning protocols, including DNA constructs and required enzymes
- Synthesizing findings across large bodies of literature to propose testable hypotheses
- Drafting experimental plans for review by experienced researchers
A researcher cited by OpenAI used the system to collect physicochemical data for 10 drugs while distinguishing predicted values from experimental measurements. The example illustrates a practical benefit: reducing manual retrieval and normalization work before scientific analysis begins. Because OpenAI selected the anecdote, independent studies will provide a stronger measure of typical gains.
Academic access broadens the evidence base
OpenAI already works with Amgen, Moderna, the Allen Institute, Thermo Fisher Scientific and Novo Nordisk on life-sciences applications. University access could produce independent evaluations, published workflows and evidence about failure modes beyond company benchmarks and commercial pilots.
The program also places GPT-Rosalind in competition with established scientific AI systems, including Google DeepMind’s AlphaFold tools. Adoption will depend on whether labs can reproduce reported gains, integrate the model with existing software and justify its cost after sponsorship ends.
Deployment details remain open
The announcement leaves several operational questions for developers and research administrators:
- Which ChatGPT, API or other interfaces are included
- What usage quotas, rate limits and seat limits apply
- How trusted-access applications are assessed and how long review takes
- What pricing or migration options apply after the sponsored period
- Which retention, security and audit controls cover sensitive datasets
Institutional review remains necessary for human genomic data, clinical records, unpublished sequences and other regulated or security-sensitive material. The training-data default addresses one concern, while each university’s privacy, biosafety, export-control and research-governance requirements still apply.
Qualified labs can find eligibility and application information on the plan page. Before building GPT-Rosalind into a research pipeline, teams will need written confirmation of interfaces, limits, data handling and post-sponsorship costs.
Comments
