OpenAI closes gap on Anthropic as AI price war escalates

The latest:

Business spending on AI models split roughly evenly between OpenAI and Anthropic in September, after Anthropic commanded three-quarters of it in January, according to an OpenRouter analysis of some 120,000 companies cited by The Wall Street Journal. OpenAI triggered the shift with cheaper models this summer. The timing matters: both companies are heading toward public listings.

Details:

• The data: OpenRouter, a startup that lets developers access different models, tracked spending by about 120,000 companies using Anthropic and OpenAI tools. The share was near parity in September versus three-quarters for Anthropic in January. OpenRouter said the data largely reflects AI-native startups, some older tech companies and large enterprises.
• The catalyst: OpenAI released its GPT-5.6 lineup — Sol, Terra and Luna — this summer, offering models with varying capabilities and lower running costs. Shortly after launch, the company cut the price of GPT-5.6 Luna by 80% and Terra by 20%, according to the report.
• A second data set: Ara Kharazian, an economist at finance startup Ramp, said data from his employer’s 70,000 customers in mid-September showed businesses spending more on OpenAI than Anthropic for the first time since December. The lead lasted days; by week’s end Anthropic was narrowly ahead again.
• The IPO clock: OpenAI said 2.5 million businesses now use its products. Anthropic is working toward a listing as soon as November, while OpenAI is likely to go public next year. Both carry skyrocketing capital expenditures against trillion-plus dollar valuations.
• The data-retention issue: Anthropic said it would retain user data for 30 days with Fable 5 for trust and safety purposes. Retool founder David Hsu said that ruled out the model for his company, which handles sensitive data. Anthropic has since given some customers control of stored data.
• Customer economics: Hsu said the GPT-5.6 release was the main catalyst for Retool’s shift, and estimated he now spends about 20% less on OpenAI models. David Zhu, CEO of sales-platform startup Reevo, cited cost and a desire to avoid depending on a single model maker.
• Anthropic’s spring surge: Demand for Claude Code was strong enough in the spring to cause frequent outages and a computing crunch. Some users later objected to guardrails placed on the Fable 5 model, released in early June.
• China factor: Companies facing mounting costs as more employees experimented with AI concluded they did not need the most advanced option for every task, and increasingly turned to low-cost, open-weight models from China for some workloads.
• Still in demand: Crowds lined up outside a downtown San Francisco warehouse on Wednesday for Anthropic’s Claude Founder House, with some turned away at capacity a day earlier. ReadyM co-founder Michael Szklarski said his startup prefers Anthropic’s Fable 5.1 but wanted to discuss lowering costs.
• The counterpunch: Anthropic began releasing its new Claude 5.5 family of models in late September, according to the report.

Between the lines:

The swing from three-quarters to parity tracks price, not capability: the customers quoted switched for cost and data-retention terms, not because they judged OpenAI’s models better. Szklarski’s startup still prefers Fable 5.1 for frontier work while seeking cheaper terms, suggesting the market is splitting between premium tasks and cheap bulk inference — a split that favors whoever prices lowest.

What’s next

Anthropic’s listing could come as soon as November, with OpenAI’s expected next year. Watch whether Claude 5.5 reverses the spending trend in OpenRouter and Ramp data, and whether OpenAI cuts prices again.