Anthropic built Fable 5 to be its best model. Two months after launch, its own customers aren’t spending like it.
Payments platform Ramp tracked spending across 70,000 companies and found Fable 5 accounts for just 11% of total spend on Anthropic’s tools. Ramp documented the plateau directly, and the number hasn’t moved since release. For a company eyeing what could be the largest IPO in history, that gap between “most capable” and “most used” is getting hard to explain away.
Enterprises Stopped Chasing the Ceiling
For most of this AI cycle, businesses grabbed whatever model sat at the top of the leaderboard. Capability mattered more than cost.
That pattern is breaking. Accel partner Miles Clements, an Anthropic investor, called the old default “not a durable era” — and the spending data backs him up. Companies aren’t rejecting Anthropic’s best work. They’re deciding they don’t need it for most tasks.
Older, cheaper models already handle the bulk of business workloads just fine. Fable 5’s price tag hasn’t bought enough everyday usefulness to pull budgets its way, and rivals selling the same pitch — Grok 4.5 undercuts Claude’s flagship pricing by design — only sharpen the comparison.
This reaches past one company’s earnings call. If frontier models become showcases instead of workhorses, the economics of AI development shift under every lab building them. Labs poured billions into bigger, more capable systems assuming capability sells itself. Ramp’s numbers say capability alone doesn’t close the sale anymore. Price and fit do.
Politics Complicated the Launch, But Wasn’t the Real Problem
Fable 5’s rollout hit turbulence early. The Trump administration forced Anthropic to pull the model days after its June debut, citing export-control concerns tied to national security, and restored access July 1.
Political risk has since faded as a factor in how companies pick models. The price-performance math hasn’t — analysts now treat it as the dominant force behind adoption, with the regulatory episode reduced to background noise. That same episode still casts a shadow in one way: the data-retention requirements attached to it have made some companies cautious about adopting Fable more broadly, according to Ramp’s chief economist, Ara Kharazian.
Opus 5 Is Quietly Winning the Internal Race
Here’s the detail that stings most: Anthropic’s own Opus 5, a smaller, cheaper model released in late July, has already overtaken Fable 5 in business spending.
That’s not outside competition eating Anthropic’s lunch. That’s Anthropic’s own customers voting between two of the company’s own products — and picking the cheaper one.
OpenAI isn’t standing still either. Its annualized revenue jumped 35% this quarter, crossing $40 billion, driven largely by GPT 5.6 — a model priced well below Fable 5. Anthropic and OpenAI are running that trade-off on fundamentally different budgets, too: OpenAI’s burn rate dwarfs Anthropic’s projected path to breakeven, which makes OpenAI’s willingness to undercut on price look more like necessity than generosity.
The Numbers Anthropic Actually Wants You to See
None of this means Anthropic is struggling. Revenue has grown nearly sevenfold since the start of the year, and Q2 brought the company’s first adjusted operating profit, with another profitable quarter expected in Q3. Anthropic now counts 6,000 customers spending $100,000 or more annually.
July’s revenue came in at $65 billion annualized, up from $47 billion in May — but short of the $80 billion some investors had penciled in. For a company heading toward a valuation north of $2 trillion, with an IPO possibly landing next month, “short of expectations” becomes a headline of its own.
Why This Should Reshape How You Think About Model Choice
This lesson isn’t specific to Anthropic. It’s a signal for anyone buying AI at scale: “biggest model wins” no longer holds.
Enterprise AI spending is starting to look like ordinary software procurement, driven by total cost of ownership rather than benchmark scores. Buyers already comparing Claude and ChatGPT on price per million tokens are running exactly the calculation this data rewards.
Kharazian put it plainly: forecasting even a few months out has turned into guesswork. Extrapolate from past trends and Anthropic should be running away with the market. Instead, a strong rival release and a lukewarm reception for its own flagship flipped that expectation on its head.
Test the cheaper model first. Increasingly, it’s already good enough.
