ChatGPT Edu Is Now Seat Plus Usage. Six Campuses Show the Enterprise Playbook

A ChatGPT Edu seat at the University of Connecticut still costs $236 a year. What changed on 1 July 2026 is what that seat includes: 60 credits a week for advanced models, with each extra credit priced at about $0.04. ChatGPT Edu pricing is now a subscription plus usage, and the public notices universities were obliged to publish are the clearest view anyone outside OpenAI has of how its enterprise seats are being restructured.

What changed in ChatGPT Edu pricing on 1 July 2026

ChatGPT Edu pricing changed from a flat seat to a seat plus metered credits for advanced features. The seat still covers base models without limit. Thinking and Pro models, Deep Research, agent mode and heavy image generation now draw on a credit allowance, and usage beyond it is either blocked or billed. UConn and Maryland both date the change to 1 July 2026.

The first public signal came from UConn's central IT team on 15 June 2026. Its notice said OpenAI had notified the university that it would move to a subscription-plus-usage pricing model in the coming fiscal year. Under the old terms, the same notice says, a set price bought access to both basic and advanced models.

Ten days later UConn published the actual terms. The base price did not move. What moved was the boundary between included and metered. That is the pattern across every campus we found, and it is the pattern worth watching in your own renewals.

The University of Maryland describes the same change in plainer terms. Its licensing site states that effective 1 July 2026, OpenAI introduced a new billing model for higher education workspaces, moving advanced capabilities to credits instead of a fixed pricing structure. Two institutions in two states, using different words, describe one vendor decision.

ChatGPT Edu pricing, campus by campus

We collected every public notice we could open that states concrete terms. Six institutions qualified. A seventh, UC Davis, published a renewal notice whose headline promises a lower cost, but its server refused our requests, so its figures are not in this table.

Every cell comes from the institution's own IT page or its student newspaper. "Not disclosed" means the source we opened did not say. Universities negotiate individually, so differences between rows are partly contract terms, not only OpenAI's list structure.

Two things stand out. First, every institution that discloses a credit unit uses one, even those that never mention the July change. Second, nobody publishes an increase in the seat price. The cost pressure arrives through the allowance, which is precisely where a budget holder is least likely to look.

The credit is the real unit of ChatGPT Edu pricing

A seat price tells you very little once usage is metered. The number that governs your bill is the price of a credit and how many credits each action consumes. OpenAI does not publish a dollar price per credit. Its pricing documentation says credit purchase prices and discounts depend on your plan or agreement. The universities, however, publish enough to reconstruct it.

Maryland's four tiers reduce to one formula

Maryland's ladder looks like four products. It is really one. Moving from Standard to Power costs $70 more and adds 1,750 credits. Power to Max costs $100 more and adds 2,500. Max to Ultra costs $200 more and adds 5,000. Every step prices the extra credits at exactly $0.04.

Run the same arithmetic backwards and the Standard seat splits cleanly. Its 500 credits are worth $20 at $0.04, which leaves $10 a month for the seat itself. On our reading, Maryland is buying a $10 platform fee plus prepaid credits, packaged as tiers. This is our inference from the published ladder, not a figure Maryland or OpenAI states.

UConn fits the same shape. Its $236 a year is about $19.67 a month. Its Base allowance of 60 credits a week, which UConn's own knowledge base restates as 260 a month, is worth about $10.40 at $0.04. That leaves roughly $9.27 a month of platform fee, close to Maryland's $10.

What 60 credits a week actually buys

Maryland publishes the rate card its users see. A reasoning message costs 10 credits, a Pro message 50, an agent message 30 and a Deep Research task 50, after small daily free allowances for image generation and Deep Research. Instant, search, GPTs, data analysis and Canvas are included without credits.

At those rates, a UConn Base seat buys about 6 reasoning messages a week, or one Deep Research task beyond the free daily quota. Washington's assumed average of 200 credits a month buys 20 reasoning messages. Those are not heavy-user numbers. They are a few serious questions a week.

Our view is that this is the design, not an oversight. The included allowance exists to make the seat feel complete on a demo. The real revenue sits in the users who discover reasoning models and come back daily, and the price sheet is built to find them.

ChatGPT Edu allowances are shrinking after the switch

A credit system has one property a flat seat lacks. The vendor can change what you get without changing what you pay. Three campuses show that lever being used within three months of the July change.

Indiana cut premium credits by about 95%

According to the Indiana Daily Student, a 1 September email from university IT moved most IU accounts from 5,000 premium credits a week to 1,000 a month. On a monthly basis, 5,000 a week is about 21,700 credits. The new allowance is under 5% of that, by our arithmetic.

The same email routes extra credits through a fiscal officer and a designated account. That is the operational signature of usage pricing. Someone in finance now approves model access, which is a cost no seat price will ever show.

The source of the cut is not stated. It may reflect OpenAI's terms, IU's own budget choices, or both. What is clear is that a central IT team decided it could not afford the old allowance once credits carried a dollar value.

Stanford's Pro accounts lost usage mid-week

Stanford's FAQ, last modified 13 October 2026, confirms that changes to Pro plans announced around OpenAI's DevDay apply to its Edu Pro accounts. Users who had already consumed half or more of their weekly allowance saw the remainder reduced to zero. University IT says it is negotiating additional credits to soften the impact.

Stanford sells that Pro seat at $51.50 a month on a contract term, and its FAQ describes it as comparable to the $100-a-month Enterprise plan. The seat price held. The allowance behind it did not. The consumer side of the same change is tracked in our timeline of every 2026 ChatGPT plan change.

Washington cannot see its own meter

The University of Washington charges $276 a year and says its contract allows an average of 200 credits per person a month. Its catalog page adds that there is currently no way for an individual to track their credit usage. Heavy users may be asked to cut back, and licences can be revoked.

Read that combination carefully. A budget defined in credits, a user who cannot see credits, and a penalty for exceeding them. We think that is the single most important warning in this whole set of notices, because enterprise rollouts are built the same way. The broader problem of credits that no user can translate into money is covered in our analysis of why credits and tokens confuse buyers.

ChatGPT Edu against ChatGPT Enterprise usage pricing

The comparison the university notices invite is with OpenAI's business plans. OpenAI publishes a ChatGPT Business price and documents flexible pricing for Enterprise and Edu together. It does not publish an Enterprise seat price.

OpenAI terms come from its pricing documentation, which we read directly, and its help centre articles on flexible pricing, which we could only read through search extracts. Treat the help centre wording as close paraphrase.

The structure is identical on both sides. OpenAI's help centre describes Enterprise and Edu workspaces buying a shared credit pool at contract level, with every seat type drawing from it for advanced features. When the pool runs out, advanced features pause unless the workspace owner enables overages or buys more. The universities are simply the customers who had to write that down in public.

The same seat-to-usage shift at OpenAI's largest rival is laid out in our analysis of Anthropic's move from seat price to usage. Two leading model vendors moving their seats in the same direction in the same year is not a coincidence of packaging. It is the economics of reasoning models reaching the invoice.

Why the education channel leads enterprise seat pricing

Education is usually the last market a software vendor squeezes. Universities buy at deep discounts, in large volumes, for reputational reasons as much as revenue. When a vendor meters that channel first, it is telling you which lever it intends to pull everywhere else.

USC's numbers show how thin the seat margin already was. In January 2026, the Daily Trojan reported that the university's subscription cost $3.1 million a year for 80,000 users, with seats negotiated down to $2.50 a user a month. That puts roughly $0.7 million, about 23% of the bill, outside the seat line, on the paper's own arithmetic.

A $2.50 seat cannot fund unlimited access to reasoning models. A credit can. So the metering of Edu reads to us less as a price rise than as OpenAI repricing the expensive part of its product separately from the cheap part.

Universities publish what enterprises sign under NDA

A public university's IT department has to explain chargebacks to departments that pay them. That forces it to publish tiers, credit values and caps. A corporate buyer signs the same structure under a confidentiality clause and never compares notes.

This is the practical reason to read campus IT pages. They are the only public record of OpenAI's per-seat contract mechanics. If your company is about to renew ChatGPT Enterprise, the UConn and Maryland pages are a better briefing than any analyst note we have seen. Comparable price movements across the wider software market are tracked in our piece on seat compression in SaaS pricing.

Where this argument is weakest

Every number above is real and sourced. The inference drawn from them is a different matter, and it has three soft spots.

Six campuses is not a market

The sample is whoever published a readable page. Institutions with good terms may have had less reason to announce anything. Universities also negotiate individually, so UConn's $0.04 is one contract's rate, not a list price. The UMD decomposition into a $10 fee plus credits is our arithmetic, and Maryland may have priced its tiers that way for its own internal reasons.

The counter-case exists too. UC Davis titled its renewal notice as offering expanded features at lower cost. We could not open it, so we cannot report its figures. If a cheaper seat with tighter credits is the common outcome, the story here is a reshuffle of value, not a squeeze.

The rate card is a moving target

OpenAI's credit rates change with its model line-up. Maryland's card lists voice at 5 credits a minute, while OpenAI's current pricing documentation says desktop voice costs 1.25 credits a minute on credit-billed workspaces. Rates can fall as well as rise, and any per-message figure in this post may be stale within a quarter.

Finally, education may not lead enterprise at all. It may simply be the first channel where OpenAI had renewal dates aligned to one fiscal year. Large enterprises on multi-year agreements could see a different structure or keep flat seats for longer. We think the direction is clear. The timing for any one company is not.

What a university ChatGPT contract should now contain

Whether you run IT for a campus or procurement for a company, the notices point to the same short list of clauses. These are the terms that would have prevented each problem above.

First, a stated dollar value per credit, fixed for the contract term. Without it, every allowance is denominated in a currency the vendor can revalue. Second, a written allowance per seat with notice before any reduction, so a Stanford-style mid-week cut cannot happen silently.

Third, user-level usage visibility as a contractual deliverable. Washington's users cannot see their meter, and a cap nobody can see is a cap nobody can manage. Fourth, rollover or at least quarterly pooling, because Maryland and Indiana both forfeit unused credits each month. A wider set of renewal protections is in our guide to AI contract clauses a CFO should insist on. The concessions worth asking for on usage terms are covered in our piece on usage-based pricing concessions.

If we were renewing a ChatGPT Enterprise contract this quarter, we would refuse any order form that states credits without a dollar price. That single gap is what turns a predictable seat budget into an open-ended one.

Frequently asked questions

Where to start this week

Pull your organisation's current ChatGPT order form and find one number: the dollar value of a credit. If it is not written down, ask your account team for it in writing before the next renewal conversation, and treat any refusal as information about where your bill is heading.

Then run one report of your heaviest 10% of users by advanced-model use. Price their last month at $0.04 a credit and 10 credits a reasoning message. That figure, not the seat price, is your real exposure under subscription plus usage.

Related reading