The Best AI Model Nobody Buys: Ramp Data Shows Fable 5 Hit an Enterprise Spending Wall

By almost every technical measure, Claude Fable 5 is the most capable AI model ever shipped. Anthropic launched it in June 2026 to near-universal acclaim, and Ramp’s lead economist Ara Kharazian calls it “the best AI model to ever hit the market.” Yet the August 2026 Ramp AI Index — published August 12 and surfacing widely this weekend after a Financial Times deep-dive — contains a statistic that should unsettle every AI lab counting on frontier-model revenue: one month after launch, businesses simply are not buying it.

The numbers

Ramp, which tracks corporate card and token spend across roughly 70,000 US businesses, found that Fable 5 accounted for just 6% of tokens businesses purchased from Anthropic, and — despite being by far the company’s most expensive model — only 11.4% of dollars spent on Anthropic models. That share has plateaued rather than climbed.

The comparison with OpenAI is brutal. GPT-5.6 Sol, OpenAI’s flagship, comprises 25% of OpenAI tokens and 23% of OpenAI spend within its customer base. Fable 5 is less popular with businesses than Sol overall: in July, it generated only about 75% as much model-attributed spend as GPT-5.6 Sol.

And inside Anthropic’s own lineup, Fable 5 has already been passed. The Financial Times reports that Opus 5 — launched July 24 at $5/$25 per million input/output tokens, half of Fable 5’s $10/$50 — has surpassed Fable 5 in business spending. Anthropic’s own launch materials claimed Opus 5 “beats Fable 5” on several internal benchmarks at lower cost per task; buyers evidently agreed, and voted with their budgets.

Model	Price (in/out per 1M tokens)	Share of vendor’s tokens	Share of vendor’s spend
Claude Fable 5	$10 / $50	6%	11.4%
Claude Opus 5	$5 / $25	surpassed Fable 5	surpassed Fable 5
GPT-5.6 Sol	~$4 / $20 (after Aug cut)	25%	23%

A price ceiling, discovered in the wild

Why does this matter beyond one product line? Because, as Kharazian puts it, Fable 5 is the most performant model on the market and the most expensive — roughly twice the price of the still highly performant GPT-5.6 Sol. When the absolute best model available stalls at this level of uptake, what you have found is an upper bound on what businesses will pay for intelligence.

“Here, more performance is not worth the price tag,” the Ramp letter concludes. For labs to reignite frontier-model adoption, they would need to demonstrate performance beyond even Fable 5 and simultaneously ensure competitors can’t come reasonably close — a combination Ramp considers “increasingly out of reach, especially as open source models catch up to being only a few months behind.”

That last clause is doing a lot of work. The share of AI-spending businesses using model-serving platforms — the gateway to open-source and Chinese-developed models — hit 6.1% in July, up from 4.5% in January. Notably, Ramp’s data shows first-time AI buyers still start with the American labs; it is the advanced, high-intensity spenders who are increasingly routing workloads to open-source alternatives. The ceiling isn’t being hit by casual users balking — it’s being hit by the most sophisticated buyers optimizing cost per task.

The wider market picture

The same index shows the competitive backdrop that makes this sting:

• Anthropic extended its lead in business AI adoption: 43.5% of US businesses paid for Anthropic subscriptions or tokens in July, up 1.1 points month-over-month.
• OpenAI underperformed overall AI adoption growth, rising just 0.23 points to 39.7% of businesses.
• xAI posted its fastest growth since July 2025, up 0.94 points to 4%.
• Overall business AI adoption on Ramp reached 55.7% — the market itself is still expanding.

And spending is not slowing in aggregate: in July the top 1% of businesses spent a median $7,400 per employee on AI annually, the top 10% spent $650, and the median firm spent $11.95. The AI economy is bifurcating into a small cadre of heavy spenders and a long tail — and it is precisely that heavy-spending cadre that is proving most price-sensitive at the frontier.

Caveats worth taking seriously

Ramp itself flags an important methodology note: the Fable usage data comes from its token spend management product, whose sample “skews slightly more tech-y” than the broader AI Index sample — meaning actual Fable 5 adoption is likely even lower than estimated. The figures measure model-attributed API/token spend and don’t fully capture ChatGPT- or Claude-style seat subscriptions, and the plateau could partly reflect Anthropic steering capacity toward its own products. But the direction of the evidence is consistent: Opus 5 overtaking Fable 5 is not a data artifact; it is a purchasing decision repeated across thousands of firms.

There is also historical irony in the details. Fable 5 is the model whose release was briefly blocked by the US government on national-security grounds earlier this summer. The model that generated the most regulatory drama of 2026 turns out to be generating remarkably little revenue drama.

What it means for the AI trade

Ramp titled the letter “Cracks in the AI Thesis,” and the argument is straightforward: if the best model ever made can’t command a premium, then the pricing power story underwriting hundreds of billions of dollars in AI infrastructure investment is weaker than assumed. Every lab now faces the same arithmetic OpenAI acted on earlier in August when it cut GPT-5.6 Sol pricing by more than 20% — compete on price-performance, not on the leaderboard.

For Anthropic specifically, the picture is not grim. The company is winning business adoption overall, and Opus 5’s cannibalization of Fable 5 may simply be a rational portfolio: enterprises buy Opus 5 for the daily grind, reserving Fable 5 for the tasks where its edge genuinely pays for itself. But that is a very different business than “the best model wins the budget,” and it prices frontier capability as a niche, not a foundation.

The lesson of the August index is one economists keep relearning: performance gets you on the shortlist; the invoice gets signed by unit economics. Fable 5 found the ceiling. The next model that wants to break through it will need to do more than score higher — it will need to cost less.