Ramp data: Fable 5, launched in early June, has plateaued at ~11% of spending on Anthropic tools, as companies shift to cheaper models; Opus 5 surpassed Fable 5
AI lab's Fable 5 has met with sluggish demand from corporate clients — Anthropic's US customers are using cheaper alternatives …
Financial TimesGeorge Hammond
Context & Ripple Effects
Fable 5 launched in June with broad availability across Pro, Max, Team, and Enterprise plans, but Anthropic set a June 22 cutoff after which the model required usage credits — a bet that businesses would pay premium per-token rates for the flagship. Ramp's July index already showed the strain: while Anthropic's overall market share hit 43.5%, widening its lead over OpenAI, Fable 5 accounted for only about 6% of the tokens businesses actually bought, attributed to high cost.
The August data confirms that was a ceiling, not an adoption curve: Fable 5 has plateaued at roughly 11% of spending on Anthropic tools, and Opus 5 has now surpassed it. The picture is a lab whose franchise is healthy but whose newest, most expensive model is being bypassed by its own customers.
First-order effects
Corporate clients using Anthropic's tools are reallocating spend toward cheaper models, leaving Fable 5 — the flagship priced behind a usage-credit wall — trailing even Opus 5 inside its own catalog.
Anthropic's revenue mix is decoupling from its frontier release cycle: the 43.5% share gain comes from cheaper tiers, not the June launch it built its marketing around.
Second-order effects
With bankers floating a potential IPO above $100B at a $2T valuation, a flagship that plateaus at ~11% of tool spending puts more weight on mid-tier volume and the usage-credit pricing model to justify that number.
Rival labs watching the same Ramp data get a clear signal that top-tier price points underperform: the competitive battleground shifts from releasing the most expensive frontier model to owning the cost-efficient workhorse tier.
Third-order effects
If buyers keep routing spend away from premium flagships, model economics invert — margins compress at the top of the line while value concentrates in cheap, high-volume tokens, pressuring labs' compute spending plans (Anthropic is already laying groundwork for its own chips) and making procurement discipline the norm for enterprise AI budgets.
Frontier launches stop functioning as revenue events and become capability proofs, with actual monetization migrating down-stack — a structural shift in how AI labs price and sequence releases.
The trend: Enterprise AI buying is entering a procurement phase where cost-efficient models beat flagship premiums, forcing labs to monetize the middle of their catalogs rather than the frontier.
Not so surprising if @AnthropicAI are struggling to attract users to Fable 5. Claude use is much more geared to coding/computing and mathematical tasks than ChatGPT, but 15%-25% of conversations still apparently search substitution. https://www.ft.com/...
More data than open-source AI is taking share from OpenAI and Anthropic. Open source has gone from 28% token share to 62% token share @vercel over the last 2 months. Chart from @rauchg Super impressive given that the sum of OpenAI and Anthropic accelerated in July. So net
“.. This breaks a pattern of corporate users defaulting to the most powerful models. .. If sustained, the shift could radically alter the business model of frontier labs ..” @FT https://www.ft.com/...
Not looking good for them: Anthropic's Fable 5 is losing ground to cheaper open-source models: According to the FT, Anthropic's most powerful model accounts for just 11% of its corporate AI spending on Ramp. The reason is simple: Fable 5 is too expensive, while cheaper models
This interesting. I stopped using Fable since I didn't see any improvement over Opus for my needs, but I didn't know that everyone else did the same. Right now I'm even considering dropping to opus 4.8, because 5 is too “pro-active”. https://www.ft.com/...
How does Ramp have good data on this? Oh that's right, they don't. A fairly sloppy reading of the data by @FT . Ramp's full customer sample (the 70,000) can tell them who Ramp customers pay and how much. The model mix data comes from Ramp's token spend management product used
The crazy part is that this isn't necessarily a “bad model” problem. If Fable 5 is genuinely strong but companies keep choosing cheaper models for most workloads, price becomes a feature. At scale, even a small difference per request adds up fast. The AI race is starting to lo…
Anthropic's US customers are using cheaper alternatives to its most powerful AI tool, raising questions about the group's high-spending and valuation - $2 trillion. https://www.ft.com/...
Disagree. Anthropic is not (yet) “cooked”. They have a decent chance of surviving. They have a lot of talent; they have significant market share. They have executed well. But anyone who invests money in them at a $2 trillion dollar valuation when interest in their premium pr…
One way to read this is “spending on Fable has plateaued” but that's not what Anthropic cares about: they care about total spend across all models! A better way to read this graph is “Fable is a great complement to the other models, which has increased overall revenue for
The best AI model doesn't always make the best business. Most companies don't need the most powerful model. They need one that is good enough for the job at the right price. AI is starting to look like a real market.
None of this is new to my Substack readers as I wrote about it three days ago with better information and details. FT: “Spending on Fable 5, Anthropic's largest and priciest model, has plateaued at only about 11 per cent of overall outlay on the company's tools, more than two
11% is the problem Anthropic can't benchmark away. More than two months after launch, Fable 5 holds only ~11% of Anthropic spend across 70,000 companies. Opus 5 already passed it. A $2T IPO story now has a pricing problem, not a capability problem.
FT reports that usage of Anthropic's Fable 5 model has plateaued. Companies are unwilling to pay the much higher cost for a slight increase in quality. China's strategy of undercutting the US AI companies with open weights models is panning out. #fable #localAI
Saying it again, for those who aren't aware: It doesn't matter if Fable makes $. You train Fable so that you can distill a better Opus off it. Any revenue that Fable makes is just a bonus.
Consensus on this seems to be ZDR which tracks to my own experience as well. If that really turns out to be the case, it's an incredibly strong and clear message to the labs on the impact of data retention policies.
Lots of folks jumping to conclusions on the FT market share data for Fable. The reason is not at all obvious to me. Is it due to cost? Speed? Aggressive moves from OAI? ZDR? Open Source? Bad press / vibes? Refusal to work in some domains? Biased data?