AI总结
- 核心结论：AI领域正爆发价格战，企业通过混用开源、国产及各类模型来降低算力成本，迫使OpenAI与Anthropic面临定价下调压力。
- 结构逻辑：
- 现象：企业因成本高昂转用廉价模型；
- 方法：动态切换模型（复杂任务用顶级，简单任务用开源/平价）；
- 影响：导致AI支出增长放缓，威胁顶尖厂商利润；
- 趋势：模型逐渐趋于商品化（Commoditization）。
- 关键信息：
- 核心趋势：AI模型“混用策略”（Mixture of models）；
- 关键数据：部分企业通过混用模型降低95%成本；DeepSeek在Vercel平台占比由1%升至17%；
- 行业相关：OpenAI, Anthropic, DeepSeek, Alibaba, Microsoft, Nvidia；
- 策略：按任务定价（Price per task）取代按Token定价。
- 关键问题与解答：
- 问题：企业如何应对不断激增的AI成本？
- 解答：利用工具在不同模型间自动切换，利用低成本开源模型执行日常工作，仅对复杂需求调用顶级算力。
正文
Startups and tech giants alike are mixing and matching AI models to avoid the premium prices charged by industry leaders
EMIL LENDOF/WSJ; GETTY IMAGES
The AI price war has begun.
Big companies and startups, chafing at rapidly escalating artificial intelligence costs, are increasingly turning to tools that tap in to cheaper AI models, including some from China. That’s raising pressure on industry leaders OpenAI and Anthropic to lower their prices, a prospect that could hurt their ability to grow into profitable enterprises.
The new cost-saving tools help businesses save on AI costs by dynamically switching among a mixture 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—including those made by Chinese companies like Alibaba and DeepSeek—for many functions. The agents only tap the most capable versions of OpenAI’s ChatGPT and Anthropic’s Claude for more complex tasks. That can reduce costs for some AI-assisted work by as much as 95%, according to executives using the tools.
“Once we find something that is working well and engineers love, we find ways to make it cost effective,” said Dan Robinson, founder of Detail, a startup that identifies bugs. “There’s really an embarrassment of riches right now coming out of the open source labs.”
Robinson shifted 90% of Detail’s workload from Claude and Google’s Gemini to custom models and GLM, a family of models developed in China.
The shift to cheaper models appears to have played a role in a recent decline in a widely followed index that tracks AI spending, Citadel Securities said in a report this week. “Even the most powerful technologies must pass through the prosaic discipline of cost curves, capacity constraints and marginal returns,” the report said.
OpenAI is considering drastic cuts to the prices it charges AI users, ahead of similar cuts the company expects at Anthropic, The Wall Street Journal reported. The company sees itself as having an advantage in such a scenario because it spent massive sums in the past year to secure access to computing resources at far lower prices than what’s available now.
Sam Altman, CEO of OpenAI, holding a phone while walking through the Hart Senate Office Building.
Sam Altman recently called costs ‘a huge issue.’ Evelyn Hockstein/Reuters
Chief Executive Sam Altman said at a recent company event that costs had suddenly become “a huge issue.”
The growing price war threatens to widen losses at OpenAI and Anthropic, which are already bleeding billions of dollars a year to pay for computing firepower to build and operate advanced AI systems. Both companies have filed confidential paperwork ahead of potential initial public offerings.
Pressure on AI prices is also a new data-point in the longstanding debate over whether lower-cost competitors will commodify AI models in coming years—or if the biggest AI companies’ fast pace of improvements will keep them ahead. Both OpenAI and Anthropic also offer cheaper models to which they can steer customers to lower costs.
Advertisement
“You don’t need a model that knows quantum gravity,” said Vishal Misra, the vice dean of computing and AI at Columbia University’s engineering school. “These open source models are very capable, and the ability to charge a big premium for AI is going to diminish.”
U.S. companies are also trying to tap in to the momentum for cheaper AI models. Microsoft unveiled a suite of smaller AI models last week it said can operate more efficiently than leading-edge models. Chip titan Nvidia has launched Nemotron, a family of cheaper models that is gaining traction, and also has backed Reflection, a startup building open-source AI.
Open-source Chinese models have been rising in popularity across American businesses. DeepSeek’s share of AI usage rose from 1% in April to 17% in May on the startup Vercel’s platform, the company said.
On OpenRouter, another startup that processes AI queries, DeepSeek has been the most-used AI company since mid-May. Among their highest-spending customers, open-source token usage grew four times faster than closed-source between fall 2025 and spring 2026, OpenRouter said. The company has also seen more than 500 organizations swap from proprietary to open-source models.
Advertisement
Anthropic CEO Dario Amodei speaking at the Code w/ Claude event.
Anthropic CEO Dario Amodei. Anthropic’s new Fable 5 model is more than 50 times more expensive than DeepSeek’s V4 Pro. Jason Henry for WSJ
Optimizing AI spending can make for complex math. Open-source models cost far less per token, the basic unit of AI computing. Anthropic’s recently-released Fable 5 model is more than 50 times more expensive per token than DeepSeek’s V4 Pro, for example.
But the top proprietary models from companies like OpenAI, Anthropic or Google remain four to six months ahead of open-source competitors, researchers say. In some cases that means they can complete a complex task using fewer tokens, equating to a lower total cost.
“Companies are increasingly evaluating models on price per task: what it costs to complete a task, start to finish, and not price per token,” an Anthropic spokesman said. The company also has lower-priced models, the spokesman said.
Advertisement
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 Lindy’s tasks of managing inboxes and calendars, drafting emails, and transcribing meetings.
They found that DeepSeek handled these tasks as well as Anthropic’s Sonnet and was good at email triaging in particular. And, Crivello said, it was 10 times cheaper.
The company still uses a more advanced Anthropic model for internal coding, but overall the move has saved the company millions of dollars, Crivello said.
Many companies have begun to design their own AI models using open-source alternatives and say they are managing to reduce AI costs. When companies build in-house models and train them with company data, their performance can improve or even exceed the capabilities of frontier AI models, executives say.
Others have begun to use tools that mix and match various AI models depending on cost and what tasks are being performed.
Liang Wenfeng, founder of DeepSeek, delivering a keynote speech at the 10th China Private Equity Golden Bull Awards.
Liang Wenfeng, founder of DeepSeek. DeepSeek was the most-used AI company in June on OpenRouter. Imago/ZUMA PRESS
“Our AIs now, they are so stingy and parsimonious,” said Andrew Moore, the former head of Google Cloud AI, whose startup Lovelace AI has a platform aimed at making AI agents more efficient. “They know exactly how to get something out of the cheapest models possible. When they get into trouble, they temporarily jump up to a higher price point with a fancier model.”
Matan Grinberg, the CEO of Factory, which offers autonomous coding tools and has developed a product that uses a mixture of AI models, said his phone has been ringing all day, every day in recent weeks, as top executives in 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 that,” said Grinberg.
News Corp, owner of The Wall Street Journal, has a content-licensing partnership with OpenAI.
Corrections & Amplifications
Qwen is an AI model produced by Alibaba that includes many open-source variations, but a newer model named Qwen3.7 is closed source. An earlier version of this article included a graphic that incorrectly labeled Qwen3.7 as an open-source model. (Corrected on June 12)
Advertisement
Appeared in the June 13, 2026, print edition as ‘OpenAI and Anthropic Are Facing Get Ready for A Price War’.