Free GPT-Rosalind for university labs: Inside OpenAI’s new academic plan

OpenAI has launched the Rosalind Academic Plan, which gives eligible academic labs free Business Standard seats for as many as 30 people, for a year. Access to its life sciences model GPT-Rosalind depends on trusted-access approval.

What the Rosalind Academic Plan includes

The plan is aimed at academic research groups working in biology and medicine. According to OpenAI, eligible labs receive three things: as many as 30 Business Standard seats priced at $0 per user per month for a maximum of 12 months; use of every commercially available frontier OpenAI model, within the plan’s usage limits; and GPT-Rosalind once trusted-access approval is granted.

OpenAI describes GPT-Rosalind as a specialized life sciences model built for deeper reasoning on complex biological questions. It is paired with Rosalind Workbench, which the company says coordinates specialized tools for work such as medicinal chemistry, genomics and help in the wet lab. The first research area the announcement names is drug and protein design, meaning work on candidate drugs, therapeutic proteins, peptides and enzymes, and on tuning their molecular interactions, selectivity, stability and development suitability.

Access is gated. OpenAI says approved researchers can use trusted access to look into sensitive scientific questions, under safeguards meant for legitimate research. The application has four steps: sign in and apply, describe the research and the lab, complete verification by sharing details and signing an agreement, and start using GPT-Rosalind once access is active. The announcement does not say how eligibility is decided, how long review takes, or how many labs will be accepted.

On data, the company states that material shared through the plan is not used by default to train its models.

How labs are already using it

OpenAI cites three academic users. The Zhao Lab at UCSF said GPT-Rosalind helped it gather drug physicochemical parameters faster, working across ten drugs with PubChem and NCBI tools while refining the distinction between predicted and experimentally measured values. Dr. Mathieu Bourdenx, who leads the Bourdenx Lab, said he uses GPT models throughout his research, from exploring the literature to refining experiments and computational approaches. OpenAI also points to work ranging from protein design to understanding neurodegeneration.

Dr. Brian Hie of Stanford University and the Arc Institute was more forward-looking. He said Rosalind will help improve current discovery and design tasks, and let researchers apply them at “larger scales than previously possible” while still giving useful insights to biological researchers.

These are statements supplied by OpenAI in its announcement. The company has not published measurements of speed-ups or accuracy for these labs.

Why it matters

The academic plan changes who can reach GPT-Rosalind. At launch, The Decoder reported that the model arrived as a research preview, and other coverage described access as limited to qualified US enterprise customers under a trusted access program. University labs were not part of that group. Free seats for up to 30 people per lab give OpenAI a way into a segment its restricted rollout had left out.

The main rival here is Anthropic, which took a different route. Its Claude for Life Sciences launch in October 2025, as Maginative described it, did not rely on a gated specialist model. It connected the general Claude model to lab and research platforms including Benchling, 10x Genomics and PubMed, and was available through Pro, Max and team plans. The two companies are therefore pursuing different bets. Anthropic builds on tool integrations around a general model, while OpenAI pairs a vetted-access model with its own Rosalind Workbench.

OpenAI’s tooling is not fully locked down either. Coverage of the launch noted a free Life Sciences research plugin for Codex that connects to more than 50 public databases and tools, including AlphaFold and BindingDB, and works with OpenAI’s mainline models, so labs without Rosalind approval can still use it. Whether the specialist model delivers results that general models plus connectors cannot is not something either company has shown publicly, and the announcement offers no head-to-head figures. For academics deciding where to build their workflows, that missing comparison matters more than the price of the seats.