OpenAI offers university labs 30 free GPT-Rosalind seats for one year – US only
On October 15, 2026, OpenAI launched the Rosalind Academic Plan, an offer that gives qualified labs at US universities up to 30 complimentary ChatGPT Business Standard seats for up to 12 months. But the model at the heart of the offer — the life-sciences model GPT-Rosalind — is not automatically included. Access is gated behind a "trusted access" review with identity verification, a requirement to name a biosafety contact, and ongoing monitoring, and the offer rules out API access entirely. That is according to Unite.AI, The Decoder, Marukei Factory and several others, all drawing on OpenAI's own announcement and help pages.
For lab leaders and research administrators, this is in practice an offer with three layers: a standard ChatGPT package that covers broadly but with limits, a highly concrete application process with multiple filtering steps, and a set of clear exclusions that determine whether the offer applies to them at all.
What labs actually get
A lab leader or designated deputy can apply on behalf of the lab. Labs that pass the review process receive up to 30 complimentary ChatGPT Business Standard seats for up to 12 months, in a single managed workspace (Unite.AI; Marukei Factory).
According to Unite.AI, citing OpenAI's help page for the program, a Standard seat includes ChatGPT and Codex — but only within the existing usage limits. The free offer does not remove those limits. That means a lab with heavy usage effectively gets the same cap as a paying Business customer.
GPT-Rosalind, OpenAI's specialized life-sciences model, is a separate question. Access to the model under the offer is conditional on "trusted access" approval — and even then it does not apply everywhere.
What the offer explicitly does not include
The list of exclusions is as significant as the offer itself:
- No API credits or API usage. Approval under the plan does not grant access to GPT-Rosalind via API, according to Unite.AI. Labs that want to build their own tools or pipelines around the model are therefore left out.
- No Zero Data Retention support. Standard data policies apply in the workspace, which may matter for labs with strict internal data-handling requirements.
- A single workspace. The offer does not cover multiple separate environments.
- US university labs only. According to Lapaas Voice, the offer does not automatically extend to commercial biotech companies, independent researchers unaffiliated with a qualifying university lab, or university labs outside the US.
That last point narrows the recipient group considerably: much of the life-sciences research in Europe and Asia, and the entire commercial sector in the US, falls outside this program.
How the approval process works
The "trusted access" review has multiple steps, according to Unite.AI's walkthrough of OpenAI's help page:
- Identity verification via Persona. The verification service Persona is used to confirm the university's and the principal investigator's (PI) identity. But completing Persona verification does not mean the application is approved — it is only one filter.
- A biosafety contact. The applicant must name a trained biosafety contact at the institution. It is a concrete requirement that reflects the fact that the model will be used in contexts where misuse is a real concern.
- Review of institution, researcher and intended use. OpenAI vets the institution, the lab's lead researcher and the planned use before access is granted, and then monitors how the model is used after approval (The Decoder, citing OpenAI).
OpenAI itself stresses, according to Unite.AI, that a university email address or an existing subscription does not guarantee approval. In other words, this is not a self-qualifying extension of ChatGPT access, but a curated process in which OpenAI retains decision-making authority lab by lab.
OpenAI frames the approved researchers themselves as people who can work on sensitive scientific questions within the safety guardrails, according to The Decoder.
Where this comes from
The Rosalind Academic Plan is not an isolated announcement, but the latest step in a series: GPT-Rosalind was launched on April 16, 2026 as a research preview, with early access limited to vetted commercial and research organizations — OpenAI has highlighted collaborations with Amgen and Moderna (Cryptobriefing). On August 28, 2026 came Rosalind Workbench, the research environment around the model, and on September 11, 2026 OpenAI announced, according to Blockchain.News, that the model had left the preview stage with global availability for qualified organizations.
The Academic Plan is thus the first time the model is opened to academic labs through sponsored seats — after roughly six months in which access was reserved for companies and research organizations that had already gone through a vetting process.
It is worth noting a discrepancy Blockchain.News has pointed out: at the same time as OpenAI announced on September 11 that the model had left the preview, it was still labeled "Research Preview" in OpenAI's own Help Center. That unresolved status picture is relevant for labs considering applying, since it is not entirely clear which product stage the offer is built on.
Data policy: a company claim, not independent confirmation
OpenAI says research data shared through the plan is not used for model training by default, and that standard data retention applies in the managed workspace (Unite.AI).
This is a company claim, not a verified third-party check. Most coverage of the launch appears to draw on the same OpenAI announcement and help pages, so the fact that multiple outlets repeat the promise does not mean it has been independently confirmed. Labs with sensitive research data should treat the promise as a contract question to examine in OpenAI's own terms, not as an established fact.
Open questions and unresolved figures
Several things remain unclear in the picture, and it is worth being honest about that:
- Conflicting benchmark figures. Cryptobriefing reports 53.7% on Genebench, 18.0% on Medchem Bench and 19.6% on Labworkbench. Blockchain.News cites a June 3 update with entirely different figures: 21.6% versus GPT-5.5's 20.4% on GeneBench, 27.5% versus 25.1% on MedChemBench, and 63.2% versus 55.8% on LabWorkBench. The figures may reflect different model versions or different points in time, but they cannot be reconciled from what is available. Neither set is independently verified; both originate from OpenAI itself via different intermediaries. We therefore present neither as established.
- API pricing is unconfirmed. Cryptobriefing reports a price of $5 per million input tokens for gpt-rosalind-discovery. No other source confirms this, and since the Academic Plan excludes API access, it is not decisive for this offer anyway.
- No primary sources among the available documents. All the coverage above is secondary reporting on OpenAI's material. Labs considering applying should verify the exact terms against OpenAI's own announcement and help pages before building plans around the offer.
- Unknown demand. It is so far unclear how much interest there will be, or how many labs will actually be approved — a figure that will determine whether 30 seats is a generous or a symbolic volume for an individual lab.
What it means for labs considering applying
For a qualifying lab, this is a concrete decision with concrete steps: the PI or designated deputy applies, the institution and applicant are verified via Persona, a trained biosafety contact is named, and OpenAI vets the institution, researcher and intended use. According to the coverage, it is labs that pass this review that receive the up to 30 free Standard seats — and GPT-Rosalind is available only to those that also receive "trusted access" approval. Even with approval, the model does not apply via API, and standard data retention still applies.
The decisive difference from commercial agreements is not the price but the control: OpenAI decides lab by lab, and keeps watching afterward. For labs comfortable with that, the offer is genuinely free capacity. For those that need API integration, data control beyond standard policies, or that sit outside the US, there is little here that changes anything.