Fable 5 has been released for over two months, yet enterprise spending remains stuck at around 11%. The Financial Times notes that this suggests frontier models may not be a rigid requirement for businesses, raising questions about Anthropic’s high-investment business model prior to its $200 billion IPO.
(Prior context: Anthropic releases Claude Opus 5! Performance approaches Fable 5 at half the price, becoming the new default model)
(Bonus context: Anthropic’s IPO scale rivals SpaceX, targeting a $2 trillion valuation)
Two months have passed since Anthropic released its most powerful and expensive model, Fable 5. In a report, Financial Times journalist George Hammond noted that after aggregating spending data from over 70,000 companies, payment platform Ramp found that Fable 5 accounts for only about 11% of Anthropic’s total tool expenditure, with nearly 90% of the budget still flowing to older, cheaper models—breaking the previous corporate habit of automatically selecting the most powerful model.
Who is actually paying for Fable?
Ramp’s sample further shows that, by cost, Fable 5 accounts for approximately 11.4% of Anthropic’s expenses; when measured in token usage—the actual amount of text processed by the model—the share drops to just 6%. Meanwhile, the smaller and lower-priced Opus 5 has surpassed Fable 5 in operational spending since its launch at the end of July.
The control group is also worth noting: In July, OpenAI launched GPT 5.6 with significantly lower pricing, boosting its annualized revenue year-to-date—calculated by multiplying monthly revenue by 12—to exceed $40 billion, a 35% increase that reversed its lagging position for most of the year.
Frontier is not a necessity
Why don’t companies necessarily use the most expensive models? Accel partner Miles Clements, who invested nearly $1 billion in Anthropic, put it bluntly: “Most people don’t need to operate at the frontier of performance.” He told the Financial Times that the phase in which customers only recognize frontier models “is not a sustainable常态.”
Clements believes that intellectual breakthroughs remain important, as they determine whether Anthropic can fulfill grand promises such as "curing diseases" and serve as a showcase to attract top researchers. However, for most enterprise customers, the most advanced models will increasingly resemble display windows rather than everyday tools.
The Financial Times reported that low-cost open-source weight models launched by China and other regions—versions in which manufacturers publicly release the model’s internal weights for external deployment and adjustment—provide customers with more alternatives, reducing the justification for businesses to use only the most expensive models.
Yet paradoxically, Anthropic’s business has not suffered as a result. This startup, led by CEO Dario Amodei, has seen its revenue grow nearly sevenfold this year and, for the first time in the second quarter, reported adjusted operating profitability—meaning its core business turned profitable after excluding certain one-time costs. According to the Financial Times, citing informed sources, Anthropic has told investors that it expects to remain profitable in the third quarter and currently serves over 6,000 enterprise clients with annual spending of more than $100,000.
The real test before a $2 trillion valuation
But whether this momentum can be sustained until the IPO remains uncertain. According to the Financial Times, Anthropic’s revenue last month reached only an annualized $65 billion, below investors’ most optimistic expectation of $80 billion. Growth slowed in June due to restrictions imposed by the Trump administration on Fable’s promotion, but has gradually stabilized in recent months.
Anthropic is preparing for an IPO expected to debut as soon as next month, with investors estimating its valuation could reach over $2 trillion, which would make it the largest initial public offering in history.
For cutting-edge labs, this is a turning point in their business model: previously, R&D budgets were poured into training larger and larger models, but now they’ve found that most customers simply want affordable, sufficient versions. When “stronger” no longer equals “worth more if it’s more expensive,” the money spent stacking computing power may only buy a valuation check that becomes harder and harder to cash.
📍Related Articles📍
Anthropic's annual revenue surges to $65 billion, leaving OpenAI's "loss-making" model behind