A major reversal is unfolding this year in the commercial competition among large AI model providers. Anthropic, once the dominant force in the enterprise market, is now facing a fierce challenge from OpenAI — and the ultimate stage for this rivalry will be the two companies' upcoming IPOs.

According to the latest data from OpenRouter, among roughly 120,000 enterprises that use both Anthropic and OpenAI products, spending share between the two companies had roughly evened out by September. At the start of the year, Anthropic alone accounted for three-quarters of total spending in that group.

The core driver of this reversal is the GPT-5.6 series that OpenAI launched this summer — a lineup designed around cost efficiency that was followed immediately by steep price cuts. At the same time, the data retention policy Anthropic applied to its flagship Fable 5 model has deterred some enterprise customers.

Both companies face sharply rising capital expenditure pressure and are both preparing for IPOs: Anthropic could go public as early as November this year, while OpenAI is expected to follow next year. Wall Street is closely examining whether the two companies can support valuations of more than $1 trillion each with sustainable business models and revenue growth.

GPT-5.6: OpenAI's Price-War Counterattack

OpenAI's counteroffensive was marked by the June launch of the GPT-5.6 series. The lineup includes three models — Sol, Terra and Luna — covering different capability tiers and giving enterprise customers more flexible cost options. Shortly after launch, OpenAI further cut the price of GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20%.

According to reports, many founders and industry analysts view the GPT-5.6 launch as a turning point in the market landscape. Peter Walker, head of insights at OpenRouter, said: "GPT-5.6 opened a new lane for OpenAI, and Anthropic's model lineup currently does not have a corresponding product of that kind."

David Hsu, founder and CEO of software development platform Retool, said his company was still switching back and forth between OpenAI and Anthropic early in the year, but the GPT-5.6 release was the "main catalyst" for Retool's shift toward primarily using OpenAI models. "We want to find the cheapest model, because the cheaper the model, the higher our revenue," Hsu said. He estimates that using OpenAI models currently costs about 20% less than Anthropic.

Fable 5 Data Policy: Anthropic's Hidden Risk

Beyond OpenAI's pricing offensive, Anthropic's own policy missteps have accelerated the loss of some customers. When Anthropic released Fable 5 in early June, it announced that it would retain user data for 30 days by default for trust and safety purposes. That policy creates compliance obstacles for industries dealing with sensitive data.

David Hsu said that default terms in almost all of Retool's contracts require immediate data deletion, and Fable 5's data retention mechanism made it impossible to meet that requirement, so the company never used the model. Anthropic has since tried to remedy the issue by giving some customers data control permissions, but the damage has already been done.

Enterprise Customers: Cost Pressure Reshapes Decision-Making

At a broader level, the decision logic of enterprise customers is undergoing a structural shift. As more employees try AI tools, cost pressure on companies keeps rising, and they are gradually realizing that not all tasks require the most advanced models. Some companies have begun incorporating low-cost Chinese open-source models into their multi-model strategies.

David Zhu, co-founder and CEO of AI sales platform startup Reevo, said his company is shifting AI usage from Anthropic to OpenAI for two reasons: cost considerations and reducing dependence on a single model vendor. "The honeymoon period of deep binding with one model vendor is over," Zhu said, though he also acknowledged that preferences could shift again at any time. "Things change too fast."

Ara Kharazian, an economist at fintech company Ramp, said he examined data from 70,000 clients of his employer in mid-September and found that enterprise spending on OpenAI models had surpassed Anthropic for the first time since December of last year. But that lead was fleeting — by the end of that week, Anthropic had narrowly reclaimed the top spot.

Anthropic: Market Heat Remains, New Models Strike Back

Despite pressure on market share, Anthropic's appeal in the developer community remains formidable. On Wednesday morning, a long line of people formed outside a warehouse in downtown San Francisco waiting to enter Anthropic's "Claude Founder House" event, with some arriving an hour early. The day before, some people waited three hours only to be turned away because the venue was packed, and the scene was compared on social media to the queues at the Coachella music festival.

Michael Szklarski, co-founder of video game startup ReadyM, said his company usually needs cutting-edge models and tends to use Anthropic's Fable 5.1, but he hopes to discuss with Anthropic engineers how to further reduce usage costs.

Under competitive pressure, Anthropic began releasing its next-generation Claude 5.5 series in late September, focusing on lower cost and higher efficiency. Leadership in AI is always fleeting, and this price war over enterprise customers is far from settled.