이미지: The Decoder
Summary
- Anthropic's flagship model Fable 5 accounted for only 6% of the company's total token sales and 11.4% of revenue in its first month after launch
- Fable 5 costs $10 per million input tokens and $50 per million output tokens, roughly twice as expensive as rivals like GPT-5.6 Sol
- According to data compiled by fintech firm Ramp, adoption growth for both OpenAI and Anthropic among U.S. enterprises is also slowing
- Fable 5 토큰 점유율
- 출시 첫 달 앤트로픽 전체 토큰의 약 6%
- Fable 5 매출 점유율
- 앤트로픽 전체 매출의 11.4%
- Fable 5 가격
- 입력 100만 토큰당 10달러, 출력 100만 토큰당 50달러
- GPT-5.6 Sol 점유율
- 오픈AI 토큰의 25%, 매출의 23%
- 매출 비교
- Fable 5는 토큰당 단가가 훨씬 높은데도 GPT-5.6 Sol 매출의 약 75% 수준
- 7월 미국 기업 도입률
- 앤트로픽 43.5%(+1.1%p), 오픈AI 39.7%(+0.23%p), xAI 4%(+0.94%p)
- 데이터 출처
- 핀테크 Ramp의 토큰 지출 관리 상품 집계, 테크 기업 비중이 다소 높음
The most expensive model is selling the least
Anthropic's flagship model, Fable 5, is widely regarded as the most powerful AI model on the market. Yet U.S. enterprises have been reluctant to pay for it. According to spending data compiled by fintech company Ramp, Fable 5 accounted for only about 6% of all tokens Anthropic sold through its API in its first month on the market. By revenue, its share was just 11.4%.
The comparison point is OpenAI's flagship model, GPT-5.6 Sol, which accounts for 25% of OpenAI's total tokens and 23% of its revenue. Ramp's analysis notes that despite Fable 5's much higher per-token price, the revenue it generated came to only about 75% of what GPT-5.6 Sol brought in.
The gap, by the numbers
| Metric | Fable 5 (Anthropic) | GPT-5.6 Sol (OpenAI) |
|---|---|---|
| Token share | 6% | 25% |
| Revenue share | 11.4% | 23% |
| Input price (per million tokens) | $10 | Relatively lower |
| Output price (per million tokens) | $50 | Relatively lower |
According to Ramp, Fable 5 is priced roughly twice as high as GPT-5.6 Sol or Anthropic's other flagship models. That said, the sample comes from Ramp's own token spend management product, which skews somewhat toward tech companies. Given that Fable 5 is mostly used for coding tasks, actual adoption could be even lower than these figures suggest.
Why enterprises aren't paying up
Ramp economist Ara Kharazian attributed Fable 5's weak adoption to its pricing, saying "the added performance doesn't justify the added cost." The implication is that a kind of ceiling has formed on how much enterprises are willing to spend on AI.
The issue may be more complicated than that, however. Even if Fable 5 outperforms previous models, if that gap isn't noticeably felt in day-to-day work, companies see no reason to pay for it. Translating generation-over-generation performance gains into a clear return on investment remains a murky exercise. This doesn't mean models at Fable 5's level represent an absolute ceiling on enterprise spending — a sufficiently superior model could still push that spending cap higher.
OpenAI and Anthropic growth also slowing
According to Ramp's July data, 43.5% of U.S. companies paid for Anthropic subscriptions or token usage, up 1.1 percentage points from the previous month. OpenAI climbed to 39.7%, but its increase of just 0.23 percentage points lagged the overall pace of AI adoption growth. By contrast, xAI grew 0.94 percentage points to reach 4%, its fastest growth rate since July 2025.
New customers are still flowing to U.S. model providers, but power users — the key customer segment that drives spending growth — are reportedly shifting toward open-source models. Ramp notes that the performance gap between open-source models and top-tier models has now narrowed to a matter of months. That, the firm says, is contributing to slowing growth for both OpenAI and Anthropic.
On August 10, Anthropic announced it would make the introductory pricing for its mid-tier model, Claude Sonnet 5 — $2 per million input tokens and $10 per million output tokens — permanent, with no expiration date. For high-volume use cases like coding agents, mid-tier model pricing directly determines operating costs. The Fable 5 case shows the opposite dynamic at the top end: pricing itself is acting as a barrier to adoption for flagship models.
What this means
The data suggests enterprises aren't ready to keep increasing AI spending indefinitely. Even when a more capable model arrives, if the improvement doesn't translate into visible value in everyday work, it won't translate into purchases either. Conversely, as open-source models close the gap with flagship models on a month-by-month basis, enterprises have growing incentive to switch to cheaper alternatives. This trend is also linked to concerns Anthropic faces ahead of its IPO over competition from low-cost Chinese models.
