Why OpenAI Might Slash Prices for Users

SPEAKER_1 (0:00)

The spirit of innovation is deeply ingrained in America, and Google is helping Americans innovate in ways both big and small.

Working with NOAA's National Hurricane Center, Google's latest AI model is providing experimental tropical cyclone simulations, bringing earlier and more accurate guidance to weather agencies and emergency service experts. This is a new era of American innovation. Learn more at g.co. slash American innovation.

Luke Vargas (0:32)

OpenAI weighs major price cuts as businesses begin to scrutinize soaring AI bills.

Plus, Canada proposes banning social media for teens. And mounting opposition to data centers prompts a rapid rethink in how the industry sources its power.

K.R. Sridar (0:49)

The utility companies try to provide them the power. But remember, it's a freeway and a highway paid for by the rate payer. And if one rate payer was consuming all that power and everybody else was paying the tolls for that road, people got upset.

Luke Vargas (1:06)

It's Thursday, June 11th. I'm Luke Vargas for the Wall Street Journal. And here is the AM Edition of What's News, the top headlines and business stories moving your world today.

The war for AI customers is leading OpenAI to consider drastic price cuts. We are exclusively reporting that the ChatGP team maker is weighing significant cuts to what it charges for tokens, the unit that AI firms use to bill for their products, anticipating that similar cuts could be coming at rival Anthropic. The price war could be an early test of both companies' revenue streams ahead of hotly anticipated public listings and comes as executives around the world begin to balk at the high prices for AI usage. Cindy Rose is the CEO of advertising giant WPP and had to say at the WSJ Leadership Institute's CEO Council Summit in London.

Cindy Rose (1:57)

In the marketing space, we're using the very expensive models because we're doing frontier reasoning, we're doing high resolution video and audio where there's no deflation, just cost increase.

So this is the problem or the sticker shock that most companies haven't yet experienced but will. I think the Uber CTO went public recently and said he burned his entire 2026 token budget in four months. Same at WPP. I've got more agents than employees now and there's a lot of unbudgeted cost associated with it. So now the pivot needs to be towards token consumption optimization, which is the next chapter.

Luke Vargas (2:34)

OpenAI confidentially filed for an IPO earlier this week and we report that CEO Sam Aldman told employees in a Slack message that the company plans to go public within the next year.

Guardrails erected around Anthropic's next-gen AI model are stirring a user backlash. Tech reporter Sam Schechner told me that it didn't take long after Tuesday's release of Claude Fable 5 for AI experts and researchers to start complaining about restrictions within the powerful model.

Sam Schechner (3:03)

The new model that Anthropic released Tuesday is an update to the Mythos model that the company first put out in April and said was too dangerous to release widely. And so what they did is they added pretty intense restrictions to what it can do to help mitigate those risks. And some users definitely reacted poorly to it.

Luke Vargas (3:25)

Sam, some of these were very, you know, noticeable. You got a pop-up saying, if you're prompting it on questions around bioweapons or cybersecurity, that said, you know, we're going to steer you now to a less capable model. Some users reported that the model is actually stricter than that. It was blocking by benign topics like math or biology or chemistry. And yet, there was a separate cohort very frustrated that some of the safeguards were not so obvious.

Sam Schechner (3:48)

Yeah, I think we've seen models sort of redirect conversations before. But what really rankled people, especially in the academic AI area and in the open-source AI field, was that Anthropic said that it would degrade the quality of its responses about high-end AI development, intentionally to make it less useful for developers looking to build AI tools. The justification for this was national security in its own terms of service, that you're not supposed to use the AI that way, and that others who are not scrupulous might build AIs without the same restrictions that Anthropic builds. But users reacted with intense frustration. They said it was gatekeeping, they were trying to harm potential competitors, and muddying just general AI research into the capability of Fable and Anthropic actually reversed part of these safeguards after the outcry. And so instead of silently degrading Fable's ability to do high-end AI research, it simply will tell users that it's not going to do that.

Luke Vargas (4:53)