A university AI tier promised privacy but not a published price

The classroom note

On 30 May 2024, OpenAI told universities it had built a version of ChatGPT for campuses. The announcement frames ChatGPT Edu as "an affordable offering for universities to responsibly bring AI to campus," powered by GPT-4o, and names prior deployments at Oxford, Wharton, the University of Texas at Austin, Arizona State University and Columbia. For a lecturer or IT director, the change was not a new chatbot; ChatGPT already existed. It was a licensed, admin-managed version, sold to the institution, with named professors already using custom GPTs built on course materials for assignments and feedback.

What the evidence says

The announcement lists what ChatGPT Edu includes: GPT-4o access, "significantly higher message limits than the free version," custom GPTs shareable inside a university workspace, over 50 languages, and "robust security, data privacy, and administrative controls such as group permissions, SSO, SCIM 1, and GPT management." It states plainly that "conversations and data are not used to train OpenAI models." No enrolment, adoption or outcome figure appears; the university quotes describe workflow uses, not measured results. The current product page, retrieved 16 September 2026, shows the same no-training-data commitment, now built on GPT-6 rather than GPT-4o, evidence that a living vendor page describes today's model, not the one a reader might assume from the 2024 date.

The implementation question

Cost is the field OpenAI did not publish. "Affordable" is a vendor's adjective, not a number; both pages route a buyer to contact sales rather than a rate card. The Enterprise and Edu help centre, as retrieved, lists admin guides for usage limits, legacy model access and workspace security, confirming a university's IT team, not an individual student, holds the controls that matter. What a campus buys is a governance layer, sign-on, permissions, a usage dashboard, on top of the model available to consumers, at a price visible only after a sales conversation.

What holds and what fails

The "not used to train models" commitment is a specific, checkable claim about data handling, and it appears unchanged across the 2024 announcement and the 2026 product page. It does not, by itself, describe how a university enforces academic-integrity policy inside a chat tool, verifies which students hold a licensed seat, or budgets for a price it has not been given in public. Treating "Edu" as a single fixed product across years is an editorial risk: the model underneath it has already changed once, from GPT-4o to GPT-6, without a change to the branding.

• What does ChatGPT Edu cost per seat at this institution, and who negotiated that figure?
• Which admin controls are actually turned on for students, versus available but unused?
• Has the underlying model changed since the licence was signed, and was that disclosed?

A named data commitment and an unnamed price sit side by side in OpenAI's own pages; a school evaluating the product has firmer ground on the first question than the second.