OpenAI offers U.S. university labs up to a year of free AI access

OpenAI is offering eligible U.S. university labs up to 30 free AI subscriptions for up to 12 months through its new Rosalind Academic Plan. Announced October 15, 2026, the program includes a route to GPT-Rosalind, its specialized life sciences model—but researchers must pass a trusted-access review before they can use it.

Free seats, two application routes

The offer covers Business Standard seats at $0 per user per month. It also includes access to all commercially available OpenAI frontier models, subject to the plan’s usage limits. GPT-Rosalind has an additional approval requirement; the free subscription offer does not remove that gate.

Principal investigators—the researchers leading a lab—and lab leads apply on their teams’ behalf. Those at universities without an OpenAI Enterprise agreement use the trusted-access application process. Labs whose universities already have an agreement must instead go through their institution’s Enterprise access process.

After approval, the lab lead can activate access for up to 30 named members. Verified leads manage permissions and are responsible for their lab’s approved use. API access—the interface that lets other software request model outputs—is not included, so the offer does not cover that route into custom research software.

From drug properties to experimental questions

OpenAI describes GPT-Rosalind as a model built for reasoning through complex biological questions. It works with Rosalind Workbench, which coordinates specialized tools across areas including medicinal chemistry, genomics and assistance with laboratory experiments. The academic plan brings the model and Workbench into the same lab offering.

The company lists several intended research areas, spanning the design of new treatments and the study of biological systems. Among them:

• Drug and protein design: developing candidate drugs, therapeutic proteins, peptides and enzymes, including work on their interactions, selectivity and stability.
• Disease detection: identifying molecular signs of disease and investigating minimally invasive tests to detect illness, track its progression and assess treatment response.
• Antimicrobial resistance: studying how microorganisms resist antibiotics, antivirals and antifungals, and developing treatments for resistant infections, biofilms and persistent disease.

Other listed areas include vaccines and immune responses, genetic and RNA therapies, agriculture, animal health, ecology and conservation. These are OpenAI’s stated research applications, rather than demonstrated outcomes from the newly announced academic plan.

OpenAI’s announcement also features accounts from academic researchers already exploring the model. One example from the Zhao Lab at UCSF concerns gathering drug-property evidence, with attention to a distinction important to interpreting the results:

GPT‑Rosalind helped me gather drug physicochemical parameters faster. PubChem and NCBI tools helped me work across ten drugs while refining the distinction between predicted and experimentally measured values

Zhao Lab, UCSF, in OpenAI’s announcement

Approval extends beyond an identity check

OpenAI says its trusted-access process checks both who is applying and what the lab intends to do. Before activation, it verifies the institution and lead’s identity, reviews the lead’s professional background and institutional appointment, and checks their authority to oversee access. It also assesses the lab’s affiliation, research activities and intended use.

The process establishes responsibilities for the lab lead and a designated institutional safety contact. Oversight continues after activation through monitoring and enforcement. For registered high-sensitivity uses, OpenAI says safety checks take the approved research context into account; it may follow up with researchers to clarify their work and confirm authorization.

OpenAI describes that approach as an effort to detect misuse while reducing unnecessary interruptions to authorized research. Separately, it says research data shared through the plan is not used to train its models by default.