Anthropic Wants You to Pay Up for Claude Fable 5

AI model developers have long offered consumers a simple deal: Use our technology for free through an online chatbot, or pay a monthly subscription to receive more usage, premium features, and advanced models. Anthropic is about to make that bargain a lot more complicated.

Starting on July 12 at 11:59PM PT, subscribers to Anthropic’s $20, $100, and $200-a-month plans will need to pay additional usage-based fees to access Claude Fable 5, the consumer version of the company’s highly capable Mythos 5 AI model. This appears to be the first time a frontier AI lab has gated a consumer AI model behind usage-based billing.

The rates will be the same as for developers using the company’s API: $10 for every million tokens sent to Claude, and $50 for every million tokens the model generates to answer your questions. So, if a subscriber to Anthropic’s $20-a-month plan sends a million tokens to Fable 5 in July, and the model uses a million tokens to answer their questions, they would owe an extra $60—or $80 total for the month. For comparison, $80 would get you about five months of Amazon Prime.

A million tokens is a lot—it’s roughly equal to 750,000 words, which is longer than the entire Lord of The Rings book series. But it’s not uncommon for AI power users to rack up thousands of dollars in API bills every month, in part because newer AI models like Fable 5 can spend a lot more tokens in a hidden chain-of-thought process to answer questions.

While “pay as you go” has long been the norm for developers accessing models through an API, AI labs have historically favored flat monthly subscriptions to generate revenue from consumers, and, in some cases, to control demand.

Still, the AI industry has been headed toward usage-based billing for a while now. Last year, AI coding startups like Cursor overhauled their unlimited AI subscriptions in favor of usage-based pricing models. And Anthropic recently started charging large business customers based on how much AI their employees used, rather than a predetermined fee. (The company could be making these changes to get its books in order ahead of a planned initial public offering.)

Some AI executives argue that subscription plans don’t make sense in the era of AI agents like Claude Code and Codex, which can use significantly more computational power than traditional chatbots.

“It’s possible that, in the current era, having an unlimited [AI] plan is like having an unlimited electricity plan,” Nick Turley, OpenAI’s former head of ChatGPT who now oversees the company’s enterprise products, said in a podcast interview earlier this year. “It just doesn’t make sense.”

Anthropic hasn’t closed the door to all-encompassing subscriptions yet. In a statement to WIRED, Anthropic spokesperson Reem Ateyeh says the company aims to return Fable 5 to Claude’s subscription plans “when sufficient capacity allows,” and intends to do so “as quickly as we can”—seemingly a reference to the company’s computational constraints. In recent years, Anthropic has struck multibillion-dollar deals for data center capacity with SpaceX, Amazon, and Google—though it’s still not as much as the company wants.

But it’s unclear when, if ever, Anthropic will not be constrained by data center capacity, and could offer Fable 5 within its subscription plans.

Will Consumers Pay Up for Claude?

The pricing change follows an extended promotional period for Fable 5, in which Anthropic offered the AI model to subscribers at no additional cost. In Anthropic’s initial June 7 blog post, the company said it expected demand for the AI model to be “very high, and difficult to predict.” Interest in Fable 5 has only grown since the US government banned it for foreign nationals, then subsequently approved it for general release on July 1.

Whether or not Anthropic frames it this way, the usage-based pricing for Claude Fable 5 is a test of consumer appetite for the company’s AI models. While Anthropic has largely focused on the enterprise market in recent years, it’s increasingly cutting into the consumer space, which has been dominated by OpenAI’s ChatGPT and Google’s Gemini.

Claude reached 245 million unique visitors in the month of May, more than doubling since February, according to a report from market intelligence firm Sensor Tower. That’s a far cry from ChatGPT’s monthly unique visitors, which Sensor Tower lists at 1.11 billion, and Google Gemini’s 662 million. But Claude is growing quickly.

Claude’s rising popularity coincides with a backlash from some technology leaders, who claim that Anthropic is charging sky-high prices while soaking up everyone's intellectual property. Anthropic is essentially betting it can become the “Apple of the AI Era,” and that there will always be a class of people willing to pay up, and play by its rules, for a premium AI model.

“The interesting dynamic happening right now is that no one wants to ask the question, ‘do I need the best intelligence for this task?’” says a person close to Anthropic, who requested anonymity to discuss the company’s thinking on this subject. “More often, when choosing what AI model to use, people are asking themselves, ‘Am I the kind of person who needs mid-level intelligence, or the best?’"

Most of the highly paid people in finance, politics, and tech that I talk to are willing to pay up for the best AI model, whether it’s at work or in their personal life. Increasingly, these folks are considering the “best AI model” to be Claude. Whether Anthropic can maintain that reputation requires it to stay ahead of the competition. (Based on the early reactions I’ve seen to OpenAI’s GPT-5.6 Sol, Anthropic has its work cut out for it.)

Companies like OpenAI and Google aren’t necessarily following Anthropic’s lead: They’re expected to introduce more ads to generate revenue in their free and low cost tiers, for example. But given how harshly Anthropic has positioned itself against advertising, raising prices may be its only solution in the near term.

In other words, the “golden era” of subsidized consumer AI subscriptions is finally coming to an end.

This is an edition of Maxwell Zeff’s Model Behavior newsletter. Read previous newsletters here.