Anthropic is taking a tough stance in negotiations with enterprise customers: once a customer exhausts the full Token allowance purchased under its contract, the discount is revoked outright. This approach hands an opening to OpenAI, which is pursuing a more flexible pricing strategy.

Managers at three software companies that buy AI services from both vendors said that although Anthropic and its main rival OpenAI both offer discounts to customers that commit to millions of dollars in annual spending, Anthropic applies an unusual rule: the discount ends as soon as a customer hits the usage ceiling set in the contract. These people said customers who reach the cap must renegotiate a new agreement or pay the higher list price.

An executive at one company said: "The moment you hit the cap, Anthropic's sales team immediately calls and emails, telling you 'we noticed you've used up your allowance; if you sign a new deal today, you can avoid overage charges.' Otherwise you're billed at list price."

Jeff Muscarella, chief innovation officer at NPI Financial, which helps Fortune 500 companies negotiate software procurement agreements, said Amazon, Microsoft and Google typically let customers keep discounted pricing on volumes above their committed purchase amounts during the term of an agreement. Software licensing consultants familiar with customers of both vendors said OpenAI's policy toward enterprise clients is more lenient than Anthropic's.

A person close to OpenAI said that when discount-agreement customers exhaust their committed allowance, OpenAI gives them the remainder of the current month plus an additional month as a window to negotiate a new contract, and only after that window closes does list pricing resume. An Anthropic spokesperson declined to comment.

Managers at three purchasers said Anthropic's model discounts run about 15% off list price. For the thousands of enterprise customers spending more than $1 million a year over the 12 months through June, that is a meaningful price reduction. Anthropic's major customers include Meta and Cursor, which use the Claude models for internal operations or to power their own products; many enterprises also buy Claude Code and Claude Cowork.

Anthropic has told at least one business partner that gross margins on its existing model sales business are about 75%. In recent months, OpenAI has been racing to catch up with Anthropic — Anthropic's revenue growth has substantially outpaced OpenAI's this year. OpenAI has publicly announced a 50% price cut on its new models for customers who access them through intermediaries such as OpenRouter and Vercel.

When OpenAI released its latest-generation model on Tuesday, CEO Sam Altman said on social media that the price per Token for the new model was halved, with even bigger cost reductions on a per-task basis. Frederik Filipsson, co-founder of software agreement negotiation consultancy Redress Compliance, said: "OpenAI is going all out to win enterprise business right now, and its discount strategy is more aggressive than Anthropic's."

Improved model quality at OpenAI has also made it more attractive to large AI buyers, such as code review service provider CodeRabbit. CodeRabbit CEO Harjot Gill said that six months ago Anthropic was the service's main model supplier; now OpenAI, whose discounts have clearly widened, has become its primary supplier. He did not disclose the specific discount level but said the three-year-old company spends tens of millions of dollars a year on AI.

These shifts come at a sensitive moment for OpenAI's sales team: the executive leading enterprise business development departed in August, and the department is undergoing a major restructuring. One employee said OpenAI is merging roles: positions previously split into pre-sales demonstrations and post-sale customer success are being combined, so that the people who demo products to customers also serve as customer success managers once a contract closes.

Gill and a manager at another software vendor who negotiates with AI labs said that as competition intensifies, both OpenAI and Anthropic have added terms to discount agreements requiring major customers to direct most of their AI procurement budgets to them — so-called share-of-wallet clauses. Licensing consultants, however, said such terms are largely on paper: OpenAI and Anthropic have no way to verify whether customers actually comply.

It is not yet clear whether OpenAI's more lenient strategy will weigh on Anthropic's growth. OpenRouter said the price cuts have indeed pushed more customers to choose OpenAI models. Still, Anthropic's revenue remains ahead, though OpenAI is also growing rapidly compared with most peers.

Because enterprises worry that the two major AI vendors may learn from their intellectual property, large companies handling sensitive business — such as Palantir, Nvidia and Booz Allen Hamilton — have imposed restrictions on model use. But multiple evaluations show that the two companies have the models best suited to automating white-collar work and scientific research tasks; for enterprises wanting to build customer service or legal automation businesses, or develop similar AI applications, it is hard to avoid them.