Anthropic Will Now Charge Subscribers Extra to Use Its Best AI Model

For years, using an AI chatbot meant picking one of two lanes: free with limits, or a flat monthly fee for more access and better models. Anthropic is now building a third lane, and it looks a lot more like a utility bill than a Netflix subscription.

Starting July 12 at 11:59 PM PT, people paying for Anthropic's $20, $100, or $200-a-month Claude plans will have to pay extra, usage-based fees to keep using Claude Fable 5, the consumer-facing version of the company's flagship Mythos 5 model. According to Wired, this is the first time a major AI lab has put a consumer model behind metered billing rather than folding it into a subscription.

The pricing mirrors what developers already pay through Anthropic's API: $10 per million tokens sent to Claude, and $50 per million tokens the model generates in response. Tokens are the chunks of text AI models read and write in, and a million of them adds up to roughly 750,000 words — longer than the entire Lord of the Rings trilogy. So the math sounds forgiving at first. But do a modest amount of back-and-forth with Fable 5 in a month, and a $20 subscriber could end up owing an extra $60, pushing their bill to $80 — enough, as Wired points out, to cover about five months of Amazon Prime.

That gap can widen fast. Newer models like Fable 5 often burn through extra tokens on hidden "chain-of-thought" reasoning before they ever produce a visible answer, and heavy users of AI tools already routinely rack up API bills in the thousands of dollars a month.

From flat fees to metered use

Charging by usage isn't new for developers building on top of AI models through an API. What's new is applying that model directly to everyday consumers who are used to picking a subscription tier and forgetting about it. Anthropic has been walking in this direction for a while: it recently shifted large business customers from set fees to pay-per-employee-usage pricing, and the broader industry has been drifting the same way. AI coding startups like Cursor ditched unlimited-usage subscriptions last year in favor of metered pricing after usage patterns made flat fees unsustainable.

Some AI executives now argue that all-you-can-use subscriptions simply don't make sense anymore, especially as AI shifts from chatbots to agents — tools like Claude Code or OpenAI's Codex that can chew through far more computing power completing multistep tasks. Nick Turley, OpenAI's former ChatGPT lead who now runs the company's enterprise products, put it this way in a podcast interview earlier this year: unlimited AI access is starting to look like "an unlimited electricity plan" — a deal that stops making economic sense once usage scales up.

Anthropic says the current arrangement isn't necessarily permanent. Spokesperson Reem Ateyeh told Wired the company intends to fold Fable 5 back into subscription plans "when sufficient capacity allows," and wants to do that "as quickly as we can." That phrasing points to the real constraint here: computing power. Anthropic has signed multibillion-dollar deals for data center capacity with SpaceX, Amazon, and Google, and by its own account, it's still not enough to meet demand.

A test of what people will pay for

The shift comes right as interest in Fable 5 is spiking. Anthropic gave subscribers free access to the model during a promotional stretch after its June 7 launch, when the company said it expected demand to be "very high, and difficult to predict." That demand only intensified after the US government briefly banned the model for foreign nationals before clearing it for general release on July 1.

Whether Anthropic frames it as a test or not, charging extra for Fable 5 effectively is one: a read on how much consumers actually value Anthropic's models compared with free or cheaper alternatives. The company has spent recent years focused mostly on enterprise customers, but it's been pushing harder into the consumer market that OpenAI's ChatGPT and Google's Gemini currently dominate. Claude pulled in 245 million unique visitors in May, according to market intelligence firm Sensor Tower — more than double its February numbers. That's real growth, but it's still dwarfed by ChatGPT's 1.11 billion monthly visitors and Gemini's 662 million.

Anthropic's bet, in effect, is that it can position itself as something like the Apple of AI: a premium product for people willing to pay for what they consider the best available intelligence, rather than chase the cheapest option. A person close to the company, granted anonymity by Wired to discuss its thinking, framed the shift in consumer behavior this way: "The interesting dynamic happening right now is that no one wants to ask the question, 'do I need the best intelligence for this task?' 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?'"

That framing carries real risk. Anthropic's reputation as the top choice for people who want the sharpest model rests on staying ahead of rivals, and that race isn't standing still — reactions to OpenAI's newer GPT-5.6 Sol model suggest Anthropic will have to keep proving it deserves the premium price tag.

Other major labs appear to be solving the monetization puzzle differently. OpenAI and Google are both expected to lean more heavily on advertising in their free and low-cost tiers rather than pushing usage-based fees onto consumers. Anthropic has been vocally critical of advertising as a business model, which leaves it with fewer obvious levers to pull. Raising prices, at least for now, looks like the path it's chosen instead.

Taken together, the move signals something bigger than one model's pricing scheme. The stretch of cheap, heavily subsidized access to frontier AI that consumers have gotten used to appears to be winding down, and Anthropic just became the first major lab willing to say so out loud with its wallet.