The bill, not the benchmark: Fable 5 took 6% of Anthropic's tokens in month one

Spending data from the financial services provider Ramp shows that in its first month after launch, Anthropic's Fable 5 accounted for only about six percent of the tokens purchased from Anthropic, and 11.4 percent of total spending on Anthropic models ^{[1] [2]}. Ramp also reports that Fable 5 brought in roughly 75 percent of the model-related revenue that OpenAI's GPT-5.6 Sol generated, despite costing significantly more per token ^[4]. This is buyer-side data rather than a vendor disclosure, which is the reason it is worth reading closely.

Start with the internal arithmetic, because it is more revealing than the headline share. Fable 5 captured 11.4 percent of Anthropic model spend on six percent of Anthropic tokens, which implies revenue per token roughly 1.9 times the Anthropic average ^[2]. GPT-5.6 Sol took 23 percent of OpenAI spending on 25 percent of OpenAI tokens ^[3], implying revenue per token slightly below the OpenAI average ^[3]. One model is priced above its own house average and is not moving volume; the other is priced at or below and is carrying a quarter of the tokens. On token share within its own vendor, Sol is running about four times Fable 5 ^[1].

The price is the stated cause. Fable 5 runs about $10 per million input tokens and $50 per million output tokens, roughly twice the cost of GPT-5.6 Sol or other Anthropic flagship models, and Ramp economist Ara Kharazian attributes the slow uptake to that price, describing it as a new ceiling on what companies will spend ^{[5] [6]}. Take the 75 percent revenue figure against a rough 2x price ratio and Fable 5's token volume implies something near 38 percent of Sol's ^[4]. That is a demand gap, not a rounding difference.

Caveats belong here. The Fable sample comes from Ramp's proprietary token spend management product and skews slightly toward tech companies, and Ramp notes actual adoption is likely lower still if Fable 5 is used mainly for coding ^[7]. The-decoder's own reading is that the story is more complicated than price alone: the performance edge may not matter for many use cases, or may be too small to measure in daily work, which is a return-on-investment measurement problem as much as a pricing one ^[14].

Breadth is still growing while depth stalls. Ramp reports 43.5 percent of U.S. companies paid for Anthropic subscriptions or tokens in July, up 1.1 points on the month, with OpenAI at 39.7 percent but up only 0.23 points, lagging overall AI adoption growth ^{[8] [9]}. xAI posted its fastest growth since July 2025, rising 0.94 points to four percent ^[10]. Ramp says heavier users are shifting toward open-source models that now trail frontier models by only a few months ^[11]. Spending itself keeps climbing: in July the top one percent of U.S. companies spent a median of $7,400 per employee on AI, the top ten percent $650, and the median company $11.95 ^[12], a spread of roughly 619 to one between the top percentile and the median ^[5].

Watch whether Fable 5's spend share rises without a price cut, and whether Anthropic cuts. Watch the monthly adoption deltas for OpenAI, which are now close to flat ^[9], and the open-model lag, because Ramp treats slowing growth at the two leading labs as a problem for an investment thesis that assumes fast-growing revenue from ever more powerful models ^[13].