Anthropic Built Its Most Powerful AI Model Yet. Businesses Aren't Buying It.

August 17, 2026

Summary: Ramp's monthly AI Index found Anthropic remains the leading AI vendor by business adoption at 43.5% of U.S. companies, but its new flagship model Fable 5 has captured only 6% of Anthropic token usage and 11.4% of Anthropic spend a month after launch, despite being priced roughly twice as high as OpenAI's leading model. Ramp's research lead argues the data suggests businesses have found an upper limit on how much they'll pay for AI performance, a signal with implications for how AI labs price future model releases.

anthropicairampaiindexfable5aiadoptionaipricingenterpriseai

Anthropic still leads every other AI company in business adoption. Its newest and most powerful model is having a hard time getting anyone to use it.

That's the headline finding from Ramp's latest AI Index, a monthly research series built on spend data from businesses using the payments platform. Anthropic captured 43.5% of U.S. business AI subscriptions in July, up 1.1 percentage points from the month before. OpenAI grew only 0.23 points to 39.7%, and xAI posted its fastest growth in a year, climbing to 4%.

A Flagship Model Businesses Aren't Rushing To

Anthropic released Fable 5 in June, a model whose launch was briefly interrupted when U.S. export-control rules temporarily suspended access before being restored in July. Ramp's data shows that a month in, Fable 5 accounts for just 6% of the tokens businesses purchase from Anthropic and 11.4% of Anthropic-related spend, despite being the company's most expensive model by a wide margin.

For comparison, OpenAI's flagship, GPT-5.6 Sol, makes up a quarter of OpenAI's token volume and 23% of spend on the platform. Fable 5 generated roughly three-quarters as much revenue for Anthropic as GPT-5.6 Sol did for OpenAI last month, despite costing around $10 per million tokens, twice GPT-5.6 Sol's price.

"More performance is not worth the price tag," wrote Ramp's Ara Kharazian, who leads the research.

Businesses Are Finding a Ceiling

Kharazian frames this as evidence of a pricing limit, not a quality problem. Fable 5 is widely regarded as the most capable model currently available. Businesses appear to be deciding that capability alone doesn't justify the premium.

That has real implications for how AI labs plan future releases. To win adoption at higher price points, a lab would need to outperform not just its own prior models but every competitor simultaneously, while open source and Chinese-developed alternatives keep closing the performance gap in a matter of months.

Ramp's data shows businesses aren't abandoning American model providers for those cheaper alternatives either. Adoption of open source and Chinese-developed models through third-party serving platforms did tick up slightly, to 6.1% of AI-using businesses. But Kharazian notes that growth is coming almost entirely from existing advanced AI spenders diversifying their usage, not from new buyers switching away from OpenAI or Anthropic.

How Much Companies Are Actually Spending

The spending data underneath all this is lopsided. In July, the top 1% of AI-spending businesses spent a median of $7,400 per employee. The top 10% spent $650 per employee. The median business across Ramp's entire dataset spent just $11.95 per employee.

Most companies, in other words, are still barely spending on AI at all. A small group of advanced adopters accounts for the overwhelming majority of enterprise AI dollars, and that group is exactly the segment now gravitating toward cheaper, open source options for at least part of its workload.

What This Means for Miami

These numbers give Miami finance and operations leaders something concrete to benchmark against. If a company's AI spend per employee is closer to the $11.95 median than the $650 mark of the top 10%, it likely hasn't scaled AI usage much beyond experimentation yet, regardless of how many tools are technically deployed.

For Miami's AI-forward startups and professional services firms, the Fable 5 data is a pricing lesson worth watching closely. Paying a premium for the most capable model on the market isn't automatically the right call, and Ramp's numbers suggest plenty of sophisticated buyers are already reaching that conclusion. Budgeting for "the best model available" and budgeting for "the model that actually gets used" may need to be two different conversations.

← Back to World AI