Financial platform Ramp published its August AI Index, according to which Anthropic's flagship model Claude Fable 5 captured only 6 percent of tokens and 11.4 percent of corporate client spending in July 2026, significantly lagging behind OpenAI's GPT-5.6 Sol, which gained 25 percent of tokens and 23 percent of spending.

What happened
Ramp recorded in July 2026 that GPT-5.6 Sol captured 25 percent of tokens and 23 percent of corporate spending, while Claude Fable 5 — only 6 percent of tokens and 11.4 percent of spending. Fable 5 costs corporations twice as much: about $10 per million input tokens and $50 per million output tokens, compared to half these prices for the competitor. At twice the cost, Fable 5 generated only about 75 percent of the revenue of GPT-5.6 Sol. Ramp analyst Ara Harrazean called this gap a new spending ceiling on AI — a price beyond which the corporate market is no longer willing to pay, even for flagship models.
Context
The AI model market has moved from a benchmark race to an assessment of real business value. Progress in open-source models has reduced the gap with the best closed products to a few months, allowing corporations to delegate routine tasks to cheaper alternatives and use premium APIs selectively. The gap in spending between the top 1 percent of companies, which spend $7,400 per employee, and the median of $11.95 shows that widespread adoption of expensive models remains out of reach for most organizations.
Why this matters for the industry
Ramp's data signals a fundamental shift: benchmarks no longer translate into corporate ROI, and vendors whose strategy is built on selling the most powerful model face a price ceiling. Advanced clients — the main growth driver for Anthropic and OpenAI — are increasingly migrating to open-source solutions, threatening the business model of closed API providers. The market is shifting toward a model-routing architecture, where premium models are applied only to tasks with measurable business gains, and mass inference moves to open-source and mid-tier models.
Why this matters for users
For those choosing an AI model for work tasks, Ramp's data shows: the most powerful model does not mean the best ROI. Companies are moving to routing — premium models for complex tasks where quality gains pay off, and budget options for routine work. If you are evaluating a model for your project, benchmarks are no longer a sufficient selection criterion, and an audit of current API spending is necessary to optimize the inference budget.
What is still unknown / limitations
Ramp's sample is skewed toward tech companies, so the actual adoption rate of Fable 5 in a broader industry is likely even lower. The data covers one month — this is a snapshot, not a long-running signal, and long-term observations are needed to confirm a sustained trend. Also, Ramp does not fully cover self-hosted and open-source solutions, which may distort the picture of overall spending distribution.
Sources
- August 2026 Ramp AI Index: Cracks in the AI Thesis
- The Decoder — Fable 5 slow adoption and corporate AI spending ceiling
- LetsDataScience — Ramp Finds Fable 5 Drew 75% of GPT-5.6 Sol Spend
Author
Look at AI, editorial team
