The AI price war has begun.
Large companies and startups, frustrated by the rapidly rising costs of artificial intelligence, are increasingly turning to tools that use cheaper AI models, including some from China. This is putting increasing pressure on industry leaders OpenAI and Anthropic to lower their prices, a prospect that could harm their ability to grow into profitable enterprises.
New cost-saving tools help businesses save on AI costs by dynamically switching between a mix of third-party AI models and in-house AI systems built using freely available, open-source models.
The ecosystem allows autonomous AI systems or agents to use cheap models for many tasks – including models made by Chinese companies like Alibaba and DeepSeek. Agents only use the most capable versions of OpenAI’s ChatGPT and Anthropic’s Cloud for more complex tasks. According to executives using the tool, this could reduce the cost of some AI-assisted tasks by up to 95%.
“Once we find something that’s working well and the engineers like it, we look for ways to make it cost-effective,” said Dan Robinson, founder of Detail, a bug-identifying startup. “The wealth coming from open source labs right now is truly an embarrassment.”
Robinson moved 90% of the detail workload to the cloud and custom models from Google’s Gemini and GLM, a family of models developed in China.
Hedge fund Citadel Securities said in a report this week that the shift to a cheaper model appears to have played a role in the recent decline in a widely followed index that tracks AI spending. “Even the most powerful technologies will have to go through cost curves, capacity constraints and the potential discipline of marginal returns,” the report said.
As The Wall Street Journal reports, OpenAI is considering a drastic cut in the prices it charges AI users, with the company expecting similar cuts at Anthropic. The company sees itself as having an advantage in such a scenario because it has spent huge sums of money in the last year ensuring access to computing resources at much lower prices than were available.
Chief Executive Sam Altman said at a recent company event that costs have suddenly become “a big issue.”
The escalating price war threatens to increase losses at OpenAI and Anthropic, which already lose billions of dollars a year paying for the computing firepower to build and operate advanced AI systems. Both companies have filed confidential paperwork ahead of a potential initial public offering.
The pressure on AI prices is also a new data-point in the long-running debate over whether low-cost competitors will market AI models in the coming years – or whether the rapid pace of improvement of the biggest AI companies will keep them ahead. Both OpenAI and Anthropic also offer affordable models that allow them to take on customers at lower costs.
“You don’t need a model that knows quantum gravity,” said Vishal Mishra, vice dean of computing and AI at Columbia University’s school of engineering. “These open source models are very capable, and the ability to charge a big premium for AI is going to diminish.”
American companies are also trying to take advantage of the speed of cheap AI models. Microsoft last week unveiled a suite of small AI models that it said could work more efficiently than leading models. Chip titan Nvidia has launched Nemotron, a family of inexpensive models that are gaining traction, and has also backed Reflection, a startup making open-source AI.
Open-source Chinese models are growing in popularity among American businesses. DeepSeek’s share of AI usage on startup Versal’s platform grew from 1% in April to 17% in May, the company said.
On OpenRouter, another startup that processes AI queries, DeepSeek has been the most-used AI company since mid-May. OpenRouter said that among their highest-spending customers, use of open-source tokens grew four times faster than closed-source between fall 2025 and spring 2026. The company has seen more than 500 organizations swap from proprietary to open-source models.
Optimizing AI spend can create complex mathematics. The open-source model costs much less per token, the basic unit of AI computing. For example, Anthropic’s recently released Fable 5 model is 50 times more expensive per token than DeepSeek’s V4 Pro.
But researchers say the top proprietary models from companies like OpenAI, Anthropic or Google remain four to six months ahead of open-source competitors. In some cases this means they can complete a complex task using fewer tokens, equating to a lower total cost.
A spokesperson for Anthropic said, “Companies are increasingly evaluating models based on value per task: the cost of completing a task, from start to finish, and not price per token.” The spokesperson said the company also has low-cost models.
AI executive assistant startup Lindy began exploring DeepSeek’s V4 model two months ago, founder Flo Crivello said. He and his 25-person team built extensive internal tooling to see if the Chinese open-source model could handle the tasks of managing Lindy’s inbox and calendar, drafting emails, and transcribing meetings.
They found that DeepSeek handled these tasks as well as Anthropic’s Sonnet and was particularly good at email triaging. And, Crivello said, it was 10 times cheaper.
Crivello said the company still uses more advanced Anthropic models for internal coding, but overall the move has saved the company millions of dollars.
Many companies have started designing their own AI models using open-source options and say they are managing to reduce AI costs. When companies build models in-house and train them with company data, their performance can outperform or even exceed the capabilities of frontier AI models, executives say.
Others have begun using tools that mix and match different AI models based on cost and what tasks are being performed.
“Our AI is now very clingy and stingy,” said Andrew Moore, the former head of Google Cloud AI, whose startup Loveless AI has a platform aimed at making AI agents more efficient. “They know exactly how to get something from the cheapest model. When they get into trouble, they temporarily move up to a higher price point with a luxurious model.”
Mattan Grinberg, CEO of Factory, which provides autonomous coding tools and has developed a product that uses a mix of AI models, said his phone has been ringing all day, every day, in recent weeks as top executives from industries ranging from finance to telecommunications have reached out to try to reduce their AI spending.
“This price war is going to be good and we want to help enable it,” Grinberg said.
News Corp., owner of The Wall Street Journal, has a content-licensing partnership with OpenAI.
