When a customer burns through the computing credits in its Anthropic contract, the phone rings quickly. The sales team calls and writes, one software executive said, to say the company has noticed the quota is gone and that signing a new deal today would avoid overage charges. If the customer hesitates, the excess is billed at list price.
The practice is at the center of a quiet split between the two leading makers of AI models. Anthropic takes a hard line on discounts, executives said, while OpenAI has chosen to be more flexible. Both companies offer discounts to customers who commit to spending at least a million dollars a year. What happens when the commitment runs out is where they part ways.
Three executives at software companies that buy from both providers described the difference. At Anthropic, a discount ends the moment a customer hits the ceiling in its contract. To keep the lower price, the customer must negotiate a new deal, and until it does, any additional use is charged at the standard rate. The sales outreach is immediate, the executives said, and the message is blunt.
At OpenAI, the clock keeps running for a while. A person close to the company said that when a customer under a discount agreement uses up its committed spend, OpenAI gives the customer the rest of the current month plus one additional month to negotiate a new contract. Only after that window closes does the price revert to the published rate.
The difference is small on paper and large in practice, consultants who advise customers on software licensing said. A company that misjudges its usage by a few days could pay thousands of dollars more under Anthropic’s approach, while an OpenAI customer would have weeks to sort out a new agreement before the price moved.
The mechanics revolve around token commitments, the contracts in which a company agrees to buy a set amount of computing or text output at a discounted rate. Annual committed spend is how both providers structure their enterprise deals, and the discount is the reward for the promise to buy in bulk. The point of dispute is what happens the day the promise is exhausted.
The split reflects different bets about how to sell. Anthropic is the challenger with fast-growing revenue and a crowded pipeline of demand, and it can afford to be firm. OpenAI, the larger incumbent, appears willing to trade a little pricing discipline for goodwill, the consultants said, at a moment when enterprises are weighing which provider to standardize on.
The stakes are rising because enterprise AI pricing has become a competitive front in its own right. As the models themselves converge in capability, the terms under which companies buy access matter more. The consultants said the discount mechanics, the overage rules and the grace periods are now part of the sales pitch, studied by procurement teams as closely as the models themselves. For customers running workloads across both providers, the rules on paper can outweigh the benchmarks in a bake-off.
The contrast shows up in how each company treats the moment of overage. One executive described the Anthropic approach as a hard stop: the sales team moves fast because the standard rate is steep, and the customer is negotiating from a position where every day of delay costs money. The OpenAI approach, by contrast, assumes the relationship will survive a few weeks of negotiation and prices the grace period into the deal.
The stakes for enterprises are growing because the bills are growing. Companies that committed to AI early are now spending millions a year on tokens, and the difference between a negotiated discount and list price can run to six figures over a quarter. Procurement teams that once treated AI as an experiment now treat it as a line item, and the terms of the contract matter as much as the quality of the model.
Analysts said the divergence is likely to persist as long as the two companies are in different positions. Anthropic, smaller and growing faster, has less reason to bend. OpenAI, with a larger customer base and a fight for enterprise share, has more to gain from flexibility. Both are watching the same accounts, and the difference in discount policy is one of the ways they compete for them. For the startups and mid-sized firms that are the most price-sensitive buyers, the split has immediate consequences: a hard stop on a discount can mean throttling usage mid-quarter, while a grace period lets the work continue while the contract is renegotiated.
Neither company has made its pricing terms public, and both declined to detail them. What the executives and consultants describe is a market still taking shape, where the rules are set deal by deal and the two leaders are moving in opposite directions. For now, the difference comes down to what happens on the day a customer runs out of tokens, and how loudly the sales team calls.
