OpenAI and Anthropic Race to Cut Enterprise Prices

The cost of frontier AI has been falling all year, and the two companies setting the pace are doing it to each other on purpose. OpenAI and Anthropic have spent 2026 cutting the price of access to their models, and the competition has grown sharper as the biggest corporate accounts come up for renewal.

The pattern started earlier in the year. The Wall Street Journal reported in June that OpenAI was weighing significant cuts to its token prices and expected Anthropic to answer in kind. Since then OpenAI has discounted aggressively, including reductions of around 50% for models reached through intermediaries such as OpenRouter and Vercel, as it works to win back enterprise share.

The fight over the fine print

The latest round is not really about the headline rate. It is about what happens when a customer runs out of discounted tokens. According to reporting from The Information, Anthropic takes a hard line: once a buyer burns through the credits in its commitment, the discount ends and the price snaps back. OpenAI, by contrast, is said to give customers the rest of the calendar month plus an extra month to renegotiate before reverting to list prices.

That sounds like accounting trivia, but for a company running large AI workloads it is the difference between a predictable bill and a nasty surprise. It is also a tell about strategy. Anthropic is protecting its margins as it prepares for the scrutiny of public markets. OpenAI is spending to grow.

Cheaper models, bigger bills

Analysts credit two forces for the broader slide in prices. One is genuine efficiency, as both companies get better at running inference for less. The other is plain competition, with Chinese open-weight models and a crowded field pushing everyone toward the floor.

Here is the twist that catches finance teams out. A lower price per token does not reliably mean a lower bill. The new generation of agents, which chew through tokens by reasoning in long chains and calling tools repeatedly, can consume far more than a simple chatbot ever did. Cut the unit price in half and let an agent work ten times as hard, and the invoice still goes up.

So the price war is real, and buyers are benefiting from it, but the savings are not landing evenly. The companies that planned for cheap AI and kept their usage in check are coming out ahead. The ones that turned agents loose and assumed falling prices would cover them are learning that the meter runs faster than the discount.