# Businesses Bought Anthropic's Cheaper Models, Not Its Best One: Fable 5 Took 6% of Tokens

Ramp's card-spend data puts Anthropic's flagship model, Fable 5, at 11.4% of dollars spent on Anthropic models, versus 23% for OpenAI's comparable tier, a gap that points to a ceiling on what businesses will pay for the last increment of capability.

- Published: 2026-08-15T06:14:29.474Z
- Canonical: https://polylog.news/ai/2026-08-15/businesses-bought-anthropic-s-cheaper-models-not-its-best-on
- Publisher: Polylog (AI desk)
- Section: markets
- Sources: [Polylog editors](https://polylog.news), [Anthropic](https://www.anthropic.com/news/claude-opus-5)

The clearest number in AI economics this week is a small one. According to the [August Ramp AI Index](https://ramp.com/data/ai-index-august-2026), Anthropic's top model, Fable 5, accounted for about 6% of the tokens American businesses bought from Anthropic over the past month, and for 11.4% of the dollars they spent there. OpenAI's flagship tier, GPT-5.6 Sol, took 25% of that company's tokens and 23% of its spend. The Russian-language channel AI ML Big Data [reported the same finding](https://t.me/ai_machinelearning_big_data/10707). It also flagged Ramp's own caveat: the panel is drawn from companies that use Ramp's corporate card and spend platform, a sample that skews toward mid-market firms and underrepresents the largest AI buyers.

Ramp's central conclusion is that businesses have hit a price ceiling for the top-tier model. Anthropic still leads on breadth of adoption: 43.5% of US businesses on the Ramp platform pay for its subscriptions or tokens, up 1.1 percentage points from the prior month, ahead of OpenAI at 39.7%. What businesses declined to do was pay roughly double the price for the best available model.

Anthropic's own product lineup reflects that pattern. It [released Claude Opus 5](https://www.anthropic.com/news/claude-opus-5) at $5 per million input tokens and $25 per million output tokens, offering near-flagship quality at a fraction of Fable 5's price. Anthropic has also repeatedly extended promotional Fable 5 access on paid plans rather than moving subscribers to full usage-based pricing.

The optimistic view still has advocates. Former Google chief executive Eric Schmidt argued this week that AI [is not in a bubble and is in fact underhyped](https://t.me/aipost/7839), because it automates accounting, billing, inventory and other routine business work. Both claims can be true at once: the workloads are real, and the market is pricing the premium tier down.

## What this means

Revenue per token, not model quality, is now the constraint binding frontier AI labs. If businesses keep concentrating volume in mid-tier models priced at a fifth of the flagship rate, the revenue curve that justifies spending on training runs and data centers flattens even as usage keeps growing. That pressures the labs, their compute suppliers, and the lenders financing new data-center capacity. Two outcomes would settle the question, and both should show up in the next few spending reports: either flagship usage rises as agentic workloads with costly errors move to the top tier, or labs keep cutting flagship prices and absorb the lost margin.

## What to watch

- Whether Anthropic converts Fable 5 promotional access into paid usage-based volume, which would show the price ceiling is negotiable rather than structural.
- The next Ramp and comparable enterprise spend panels, especially whether cheaper open-weight models take share from paid application programming interfaces.
- Whether OpenAI or Anthropic reprices its top tier downward, which would confirm that the ceiling, not capability, sets the market.