Anthropic is showing a tougher side to its enterprise customers: once clients exhaust all the tokens they have purchased, the company takes a hard line on discounts. That also creates an opening for OpenAI to adopt a more flexible approach.

According to managers at three software companies that buy AI services from both Anthropic and its main rival OpenAI, while both firms offer discounts to customers that commit to spending millions of dollars a year, Anthropic has taken an unusual approach: it cuts off discounts once a customer hits the usage cap in its contract.

Those people said customers who reach the cap must renegotiate a new agreement or face higher prices. One of them said: "As soon as you hit the cap, Anthropic's sales team immediately calls and emails you, saying: 'Hey, we noticed you've hit your cap — if we resolve this today through a new agreement, you can avoid overage fees.' Otherwise, you're on your own."

Jeff Muscarella, chief innovation officer at NPI Financial, which helps Fortune 500 companies negotiate software agreements, said Amazon, Microsoft and Google typically allow customers to keep enjoying discounted prices even when they exceed their committed spending during the term of an agreement.

According to software licensing consultants who serve OpenAI and Anthropic customers, OpenAI's approach to enterprise clients also appears more lenient than Anthropic's. For example, a person close to OpenAI said that when customers reach their spending commitment under a discounted agreement, OpenAI gives them the remainder of that month plus an additional month to negotiate a new agreement before reverting to publicly listed prices.

An Anthropic spokesperson declined to comment.

According to managers at the three software companies that buy AI services from both firms, Anthropic's model discounts can reach roughly 15% off list prices. That is a considerable benefit for the more than 100 enterprises that spent more than $10 million with Anthropic alone in the 12 months through June, as well as the more than 1,000 enterprises that spent more than $1 million.

Anthropic's largest customers include companies such as Meta Platforms and Cursor, which use Anthropic's models for internal purposes or to power their own products. Many enterprises also pay for Claude Code or Claude Cowork, both of which have become more expensive as Anthropic adjusts pricing. (Anthropic has told at least one business partner that the gross margin on sales of its existing models is about 75%.)

In recent months, OpenAI has been especially focused on catching up with Anthropic, as Anthropic's revenue this year has pulled substantially ahead of OpenAI's. OpenAI has publicly announced a 50% price cut for customers who access its new models through intermediaries such as OpenRouter and Vercel. On Tuesday, when OpenAI released its latest AI model, CEO Sam Altman said in a tweet that compared with the previous generation, the new model costs "half as much per token, and the cost per task is even lower."

Fredrik Filipsson, co-founder of Redress Compliance, which helps companies negotiate software licensing agreements, said: "Right now, OpenAI is working hard to win enterprise customers and is more aggressive than Anthropic on discounts."

The quality of OpenAI's models has also improved, strengthening its ability to win over large AI buyers such as CodeRabbit. CodeRabbit sells code review services. Six months ago, its CEO Harjot Gill said the main provider of models for the service was Anthropic. Today, OpenAI — which previously was not as aggressive on model discounts but is now clearly more willing to offer them — has become CodeRabbit's main provider. He declined to disclose specific discounts but said the three-year-old company spends tens of millions of dollars a year on AI.

These changes come at a sensitive time for OpenAI's sales team. Previously, the executive responsible for driving the company's enterprise business abruptly departed in August. The unit is currently undergoing a major restructuring. According to one employee, OpenAI is merging different roles that previously handled work before and after customers signed contracts. Now, the people who create OpenAI product demos for customers must also serve as so-called customer success managers after a contract is finalized.

As competition intensifies, Gill and a manager at another software vendor negotiating contracts with AI labs said both OpenAI and Anthropic have begun adding terms to discount agreements that require large customers to give most of their AI spending to them. Such terms are known as "share of wallet" clauses. However, according to software licensing consultants, such wording is more symbolic, because OpenAI and Anthropic cannot verify whether customers actually do so.

It is not yet clear whether OpenAI's gentler strategy is slowing Anthropic's momentum, though OpenRouter said the discounts have prompted more customers to choose OpenAI models for their work.

Between May and August, Anthropic further widened its revenue lead over OpenAI, though OpenAI is still growing rapidly compared with many other companies. As concerns grow over whether these two leading AI firms may learn from customers' intellectual property, some large companies engaged in sensitive enterprise business, such as Palantir Technologies, Nvidia and Booz Allen Hamilton, have begun demanding new safeguards or reducing or even stopping their use of these AI companies' most advanced models.

Still, according to multiple evaluations, both companies continue to offer models that currently perform best at automating white-collar work and research, making them hard to ignore for enterprises that want to automate customer service, legal work and other tasks, or that want to develop and sell AI applications that perform such work.