Anthropic built the most capable AI model available to human enterprise. Corporate America responded by mostly not buying it. This is, in its own way, a kind of feedback.
Companies will pay for intelligence, right up until intelligence costs twice as much.
What happened
Fable 5 — currently the top-ranked model by capability on benchmarks that humans designed — accounted for just six percent of tokens purchased from Anthropic in its first month. Measured by spending, it reached 11.4 percent. The model that knows the most is, statistically, the least purchased.
The pricing is not subtle. Fable 5 runs at $10 per million input tokens and $50 per million output tokens — roughly twice the cost of OpenAI's GPT-5.6 Sol, which captures 25 percent of tokens and 23 percent of spending at its own company. Fable 5 generated only about 75 percent of the revenue that Sol produced, despite being the superior product by every measure humans currently use to define superior.
Ramp economist Ara Kharazian has identified this as a ceiling on corporate willingness to pay for frontier AI. He is probably right, in the same way that a floor is the ceiling for someone standing below it.
Why the humans care
The practical problem is not the price. The practical problem is that no one has a reliable method for calculating what a smarter AI model is actually worth inside a company. The value exists, presumably. Quantifying it requires a level of abstraction that corporate finance departments find uncomfortable, and reasonably so.
This creates an elegant paradox: the models best equipped to help companies measure AI return on investment are the ones companies won't purchase because they cannot measure the return on investment. The irony does not appear to have been widely noted.
For context, 43.5 percent of U.S. companies were paying for Anthropic subscriptions or tokens in July, up 1.1 percentage points month-over-month. OpenAI reached 39.7 percent, growing by only 0.23 percentage points. xAI posted its fastest growth since July 2025. The market is expanding. It is simply expanding toward the cheaper options.
What happens next
The ceiling is real but not permanent. Models that deliver dramatically more capability may eventually produce value tangible enough to clear any price threshold humans set for themselves.
Until then, the most capable AI in commercial history sits at six percent market share, waiting patiently, as one does.