OpenAI is gaining ground on Anthropic as the AI price war heats up. Data from OpenRouter showed spending between the two giants became even in September. David Zhu, CEO of Reevo, said, “The honeymoon phase of being tied to one model maker is gone.” Both companies now target public offerings.
OpenAI stoked the price war this summer with the release of its GPT-5.6 lineup—Sol, Terra and Luna—giving customers access to models with varying capabilities, including some that are less expensive to run. Its Codex coding product and other business tools are now gaining ground, and it has released additional iterations of its cost-efficient models.
An analysis from OpenRouter, a startup that allows developers to access different models, found that among some 120,000 companies that use Anthropic and OpenAI tools, the share of spending was roughly even between the two AI giants in September. At the beginning of the year, Anthropic commanded three-quarters of that total. OpenRouter said its data largely represents spending by AI-native startups as well as some slightly older tech companies and large enterprises.
OpenAI, which said 2.5 million businesses now use its products, is narrowing the gap in the battle for business customers at a crucial time, as both model makers are planning initial public offerings. Anthropic is working toward an IPO as soon as November while OpenAI is likely to go public next year.
Both AI companies have skyrocketing capital expenditures, and the coming months are critical in showing Wall Street they have sustainable businesses and revenue streams to back up trillion-plus dollar valuations.
David Zhu, co-founder and chief executive officer of AI sales-platform startup Reevo, said his company’s AI use is shifting away from Anthropic to OpenAI models for two reasons: cost and a desire to be less reliant on any one model maker. “The honeymoon phase of being tied to one model maker is gone,” said Zhu. He said his preference could shift again. “Things change so quickly,” said Zhu.
On Wednesday morning, a sea of people snaked around a downtown warehouse in San Francisco, waiting to get into the hottest event in town: the Claude Founder House, hosted by Anthropic. Founders and developers started lining up an hour before the event’s start time.
An event staffer at one point shouted, “You need to have a ticket. If you are on the wait list, we will not be approving you.” A day earlier, some people waited for three hours, only to be turned away because the networking event reached capacity, with the large crowd drawing comparisons on social media to lines outside the Coachella music festival.
Michael Szklarski, co-founder of videogaming startup ReadyM, who waited in line for Claude Founder House on Wednesday, said his company often needs the most cutting-edge models and prefers Anthropic’s Fable 5.1 model. But he hoped to meet with Anthropic engineers to discuss, among other things, how to bring down the cost.
Ara Kharazian, an economist at finance startup Ramp, said he looked at data from his employer’s 70,000 customers in mid-September and observed that businesses were spending more on OpenAI’s models than Anthropic’s for the first time since December. But the lead was short-lived. By the end of the week, Anthropic was back on top by a slim margin. Ramp said its data set skews toward high-growth, tech-forward companies of varying sizes, with some Fortune 500 companies included. Its data omits spending by individual developers.
Many founders and industry analysts point to OpenAI’s June release of the GPT-5.6 models—designed with cost efficiency in mind—as a turning point.
Shortly after the models launched, OpenAI lowered the price of GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20%.
“The GPT-5.6 opened up this new lane for OpenAI in a way that the Anthropic family of models doesn’t really have,” said Peter Walker, head of insights at OpenRouter, which is owned by fintech company Stripe.
David Hsu, founder and CEO of software-development platform Retool, said that at the start of the year, his company moved back and forth between models from OpenAI and Anthropic. But the release of GPT-5.6 models was “the main catalyst” for why Retool mostly uses OpenAI models now, he said. “We want to find the cheapest models because our revenue goes up the cheaper the models are,” said Hsu.
Today, he estimates that he spends about 20% less using OpenAI’s models than models from Anthropic.
Another factor in Hsu’s shift was Anthropic’s announcement that with Fable 5, the company would keep data from users for 30 days for what it said were trust and safety purposes. That was a problem for companies working in industries that handled sensitive data—including Retool. Anthropic has since tried to address the issue with some customers by giving them control of the stored data.
“Pretty much all the contracts we’ve signed, the default is all data is deleted,” said Hsu. “But we could not attest to that if we had used Fable, so we just never used it.”
But any lead in AI can change. In late September, Anthropic started releasing its new Claude 5.5 family of models. The benefits it touted: lower cost and better efficiency.