Anthropic just watched its own customers walk away from its most expensive product, and business kept growing anyway.
The company sells two flagship style models. Fable 5 is built for tasks that run on their own for days, the kind of work where an AI has to stay on track through many connected steps without a person checking in. Opus 5 launched on July 24, 2026, priced at half of Fable 5's rate, and aimed at everyday coding and office tasks.
Within roughly a month, spend tracking firm Ramp found that Opus 5 had overtaken Fable 5 in corporate spending on Anthropic's models. Fable 5's share of tracked Anthropic spending had leveled off around 11 percent, even though it carries a much higher price and was marketed as Anthropic's strongest generally available system.
A second data source tells a different story. Vercel, which routes AI traffic for many companies through its own service, measured Fable 5 at over 13 percent of all AI spending on its platform in July, trailing only Anthropic's previous flagship. Nine out of ten teams using Fable 5 that month had never used it before. Two trackers, two conclusions. Neither one covers the whole market, and that gap is the real lesson here: nobody, including the AI companies themselves, has a full picture of how businesses are actually using these tools.
What both data sets agree on is the mechanism. Moving work from one AI model to another now takes a single line of code on many platforms. There is no retraining, no new contract, no waiting period. That kind of low friction turns model choice into a dispatch decision made fresh for every task, not a long term vendor relationship. Send routine work to the cheap model. Save the expensive one for jobs where a mistake is costly or the task needs to run unsupervised for a long stretch.
This should worry any AI vendor charging a premium for being the smartest option on paper. Once a cheaper model is good enough for most of the work in front of it, being the best no longer guarantees being the most used. Price per finished task, not price per unit of text, decides the winner.
There is a practical catch for anyone paying a flat monthly subscription instead of a per use bill. Anthropic shows individual subscribers a progress bar and a reset countdown, but never says how many units that bar actually represents. You can watch your allowance drain without ever knowing what it was worth. If your business pays for any AI tool by subscription rather than by usage, ask the vendor for that number directly, because judging value without it is close to guesswork.
None of this means Anthropic is struggling. Its revenue pace reached 65 billion dollars a year in July, up sharply from two months earlier. The company is not losing customers, it is watching them redistribute spending inside its own lineup toward the cheaper option, which is a very different problem to have.