The Best Model on the Market Is the One Companies Are Not Buying
Spending data from Ramp shows Anthropic's Fable 5 taking just six percent of the tokens companies buy from Anthropic. The most capable model is losing on price, and that says something about the whole industry.
Anthropic’s Fable 5 is widely treated as the most capable AI model available. According to spending data from the financial services company Ramp, American businesses are barely buying it. In its first month after launch, Fable 5 accounted for about six percent of the tokens companies purchased from Anthropic, and 11.4 percent of what they spent there. For comparison, OpenAI’s GPT-5.6 Sol takes 25 percent of tokens and 23 percent of spending at OpenAI. Despite costing considerably more per token, Fable 5 brought in only about 75 percent of the revenue GPT-5.6 Sol did.
The obvious explanation is price. Fable 5 runs roughly $10 per million input tokens and $50 per million output tokens, about twice what GPT-5.6 Sol or Anthropic’s other flagships cost. Ramp economist Ara Kharazian reads this as a ceiling on what companies are willing to pay: the extra capability is not worth the extra money to them. It is probably more complicated than that. For a lot of everyday work the performance gap simply does not show up, and even when it does, nobody has a good way to measure it. Putting a euro figure on the difference between one model generation and the next is genuinely hard, and finance departments do not approve things they cannot measure.
A note on the data before drawing conclusions from it: Ramp’s sample comes from its own token spend management product and skews slightly toward tech companies. Ramp itself thinks real Fable 5 adoption may be lower still, on the assumption that it is used mostly for coding.
What is actually going on here
The rest of Ramp’s August index fills in the picture. In July, 43.5 percent of US companies paid for Anthropic subscriptions or tokens, up 1.1 points, and Anthropic has now passed OpenAI, which sits at 39.7 percent and grew by only 0.23 points. xAI grew fastest in relative terms, to 4 percent. Total spending keeps climbing: the top one percent of companies spent a median of $7,400 per employee on AI in July, while the median company spent $11.95. But Ramp also sees advanced users, the ones whose growing bills the big labs depend on, shifting toward open models, which now trail the frontier by only a few months. That is the uncomfortable part for an industry whose investment case assumes revenue rises with capability. It suggests companies will pay for AI, but not automatically more for better AI.
What this means for you: if you pay for an AI subscription, the finding travels down to your level intact. The most expensive tier is not automatically the right one, and the honest test is whether you can notice the difference on your own work rather than whether it wins benchmarks. Try the cheaper model on a week of real tasks. If you cannot tell, you have your answer. If you are choosing tools for a small business, the same logic applies with more money attached, and open models are worth a look before you commit.
Sources
Google's Gemini 3.7 Flash Is Better at Code and Costs Half as Much
Three weeks after Gemini 3.6 Flash, Google shipped 3.7 Flash with real coding gains and cut the launch price in half. The catch is in the calendar: that price expires on December 31.
