Cost comes into focus again as enterprise customers scrutinize AI spending
issued Mon, Jul 13, 2026, 09:57 AM
[NEW YORK] Three prominent artificial intelligence developers have released new models in the past week. They all promise more sophistication, but their biggest selling point for now may not be what they can do, but how cheaply they can do it.
OpenAI said its most advanced product, GPT-5.6, is designed to get more work done while using significantly fewer tokens, the units of data processed by AI models. This makes the software much more cost-effective for customers.
Elon Musk’s SpaceXAI’s Grok 4.5 is touted to be twice as token efficient as comparable models from other companies.
Meta Platforms also has very “competitive” pricing for Muse Spark 1.1, CEO Mark Zuckerberg told Bloomberg.
The renewed focus on cost coincides with enterprise customers scrutinizing their AI spending.
At the beginning of 2026, companies encouraged their employees to outdo each other by leveraging AI as much as possible. This is a practice known as tokenmaxxing.
But in recent months, some companies have imposed tougher restrictions after experiencing sticker shock. This is partly because developers like Anthropic have switched to pay-as-you-go pricing rather than simply charging a flat subscription fee.
Gautier Cloix, CEO of Paris-based AI startup H Company, said he has spoken to many executives who have inflated their bills by using OpenAI and Anthropic’s models.
Croix said one CEO showed him an invoice showing that one month’s use of an AI model would cost millions of dollars.
Related items
“Companies are spending a lot more than they used to,” said Gil Luria, director of technology research at DA Davidson & Company. “As they see these costs spiraling out of control, they’re starting to question efficiency.”
As a result, top AI developers must find ways to maximize value for current and potential customers without compromising themselves to the point where they struggle to recoup the hundreds of billions of dollars invested in chips and data centers.
Zuckerberg said Meta benefits from a lucrative online advertising business and is prepared to be “aggressive.”
“The pricing of some of the other labs is very extreme and the profit margins are very high,” he said in an interview. “We think we have a real ability to deliver frontier or very advanced intelligence at a much more affordable cost.”
OpenAI may have less room for change, but we also believe it needs to be cost competitive. “Every company right now is thinking about spending and the value they’re getting in exchange for AI. This is what we really want to do,” OpenAI CEO Sam Altman said in an interview with CNBC on Thursday, July 9.
This rhetoric is a marked departure from a year or so ago, when OpenAI executives were publicly talking about charging thousands of dollars for monthly subscriptions to its top-tier AI models to better reflect the increased value it provides to businesses.
Now, in addition to introducing new, more efficient models, OpenAI is taking steps to help businesses manage their AI spending. Last month, the ChatGPT maker introduced credit usage analytics and updated spending management.
As companies become more cost-conscious, they are “looking for other solutions,” Luria said. And there’s no shortage of alternatives.
Chinese tech companies like DeepSeek are flooding the market with more affordable open AI models. Although these services still lag behind the most advanced options from American companies, the software is sufficient for many day-to-day tasks.
To secure better prices, some users are turning to model routing services that allow them to seamlessly choose from hundreds of AI models for different tasks. One such service, OpenRouter, raised over $100 million in funding in May to meet demand.
By focusing on cost efficiency, AI developers could put more pressure on Anthropic, which many see as the current front-runner.
Anthropic’s Opus and Fable models rank among the most expensive based on cost per task, according to data from benchmarking service Artificial Analysis.
Musk took particular aim at Anthropic this week in a post promoting Grok 4.5. “This is an Opus-class model, but faster, more token efficient, and lower cost,” he wrote on social media. bloomberg
