Anthropic’s most powerful AI model is facing an uncomfortable problem: businesses like what it can do, but many are not willing to pay its premium price.
Claude Fable 5, the company’s flagship model and one of the most capable AI systems available, accounted for just 11.4% of dollars businesses spent on Anthropic models during its first month of measurable adoption, according to spending data from Ramp. It represented only 6% of the tokens purchased from Anthropic during the period.
The data offers a glimpse into a potentially important shift in the AI market. Companies that once appeared eager to use the most advanced model available are increasingly asking a more practical question: Does the extra intelligence justify the extra cost?
For Anthropic, that question arrives at a sensitive moment. The company is growing rapidly and preparing for a potential blockbuster public offering, but the lukewarm corporate appetite for its most expensive model could challenge the economics of continually building ever-larger AI systems.
Why are businesses avoiding Anthropic’s most powerful AI model?
The simplest answer is cost.
Ramp’s August AI Index found that Fable 5 costs roughly $10 per million tokens, making it about twice as expensive as OpenAI’s GPT-5.6 Sol. Despite its performance advantage on some evaluations, Fable 5 captured a much smaller share of business usage than its price might suggest.
For companies operating AI at scale, the difference can become enormous.
A business using an AI model for thousands or millions of tasks does not necessarily need the absolute best model for every request. If a cheaper model can draft an email, summarize a report, classify documents or write routine code adequately, paying a premium for frontier-level reasoning may simply add to the bill without producing enough additional value.
That is changing the way corporate buyers think about AI.
What does Ramp’s data show about Fable 5?
Ramp analyzed spending data from businesses using its financial platform. Its August 12 report said Fable 5 accounted for 6% of tokens purchased from Anthropic and 11.4% of spending on Anthropic models during the month following its release.
The company cautions that its sample skews somewhat toward technology-orientated businesses and says actual Fable 5 adoption could be even lower.
The comparison with OpenAI is particularly revealing.
Ramp said GPT-5.6 Sol represented 25% of OpenAI tokens and 23% of OpenAI spending in the comparable period. Fable 5 generated about 75% as much model-attributed business spending as GPT-5.6 Sol during July.
That suggests corporate buyers are not simply chasing whichever model has the strongest benchmark scores.
They are weighing performance against price.
Does this mean businesses do not care about AI performance?
Not at all.
Businesses still want highly capable AI systems, particularly for complex coding, research, reasoning and other tasks where mistakes can be expensive.
The problem for frontier labs is that the value curve may be flattening.
A model that is 10% or 20% better at a difficult task may not be worth twice the price if a less expensive model already performs well enough for most of the company’s workload.
Ramp economist Ara Kharazian described the result as a possible ceiling on what businesses are currently willing to spend for additional AI performance.
That is a critical distinction. The market is not necessarily rejecting better AI. It may be rejecting the idea that every improvement should command a dramatic price premium.
Why are cheaper AI model gaining ground?
The market now has more alternatives than it did a year or two ago.
Open-source and open-weight models have become more capable, while Chinese AI developers have introduced systems that can compete with leading US models on a growing range of tasks.
Ramp found that 6.1% of businesses using AI were using model-serving platforms that provide access to open-source and some Chinese-developed models in July, up from the previous month.
That share remains relatively small, but the direction matters.
Companies are gaining more bargaining power because they no longer have to choose between a handful of expensive frontier systems. They can mix models, run smaller systems for routine work and reserve expensive models for tasks where additional reasoning genuinely matters.
The result is a more price-sensitive AI market.
Is OpenAI benefiting from Anthropic’s pricing problem?
There are signs that OpenAI has regained momentum.
Ramp’s data showed Anthropic remained the leading AI provider by the percentage of US businesses paying for its subscriptions or tokens in July, at 43.5%, compared with 39.7% for OpenAI. But OpenAI’s GPT-5.6 Sol has been gaining attention because of its combination of capability and price.
OpenAI itself has emphasized that GPT-5.6 is designed around “performance per dollar,” rather than simply maximizing benchmark performance. The company says GPT-5.6 Sol can deliver strong performance at lower estimated cost than competing frontier models, including Fable 5.
OpenAI also cut GPT-5.6 Sol’s API and credit pricing by more than 20% for three months beginning August 21, adding another layer to the price competition.
That creates a difficult environment for Anthropic.
If competitors can offer near-frontier performance for substantially less, the best model in absolute terms may not be the best-selling model.
What happened when the US government temporarily restricted Fable 5?
Fable 5’s launch was also unusually complicated.
Anthropic introduced Fable 5 on June 9, describing it as its most capable generally available model. Just days later, on June 12, the US government imposed export controls affecting Fable 5 and Mythos 5. Anthropic temporarily suspended access to both models while it worked out how to comply.
The restrictions were lifted at the end of June, and Anthropic restored global access to Fable 5 beginning July 1.
That interruption was clearly capable of affecting adoption.
But the current evidence suggests political uncertainty is not the whole explanation for the weak corporate uptake. The stronger issue is economics: businesses appear to be deciding that the premium model is not necessary for enough of their workloads to justify widespread use.
Why does this matter for Anthropic’s business model?
Training frontier AI model requires enormous amounts of computing power, data, engineering talent and infrastructure.
That creates an obvious business challenge.
If companies respond to each new generation by buying the most capable model at a premium, AI labs can potentially generate enough revenue to support another expensive generation.
But if customers increasingly choose cheaper models, the relationship between model capability and revenue becomes less straightforward.
The laboratory can spend billions making a system smarter without necessarily generating billions in additional customer spending.
That is the economic tension at the center of the current AI race.
Does cheaper AI threaten frontier AI model development?
It could change how frontier development is financed, although it is too early to say that the model is broken.
There will always be customers for extremely capable AI in areas where better reasoning can produce substantial economic value. Drug discovery, advanced software engineering, scientific research and sophisticated autonomous systems are examples where incremental performance may be worth a great deal.
But businesses also need thousands of ordinary tasks completed cheaply.
The likely result is a more layered AI market rather than a single winner.
Premium frontier models could become specialist tools used for the hardest problems, while smaller and less expensive systems handle the bulk of day-to-day corporate workloads.
That would still make frontier models commercially important. It would simply mean that they are used more selectively.
Is Fable 5 actually underperforming?
The answer depends on what “underperforming” means.
Anthropic launched Fable 5 as its most advanced generally available model, and its own testing showed strong performance across software engineering, knowledge work, vision and scientific research.
Ramp’s data does not show that the model is technically poor.
It shows that businesses are using it less than its premium positioning might imply.
That distinction is crucial.
Fable 5 can be an excellent model and still be a difficult product to sell at scale if customers believe cheaper systems deliver sufficient results.
In other words, the challenge may not be intelligence. It may be unit economics.
What does the rise of cheaper models mean for AI companies?
The AI industry may be entering a new phase in which efficiency matters as much as raw capability.
The first stage of the AI race was largely about building bigger and smarter models.
The next stage could be about making those models economically useful.
That means lowering inference costs, improving token efficiency, selecting the right model for each task and giving customers better ways to control spending.
OpenAI is already emphasizing those ideas with GPT-5.6, while Anthropic’s own product line offers multiple tiers of capability. The fact that Anthropic’s cheaper Opus model can compete for corporate spending with its much more expensive flagship reinforces the point that customers do not necessarily want maximum capability all the time.
What does this mean for the AI IPO race?
The timing is particularly significant because Anthropic is reportedly preparing for a potential initial public offering.
Investors will be watching revenue growth closely, but they will also want to understand how sustainable that growth is.
A company can grow rapidly while still facing difficult questions about margins if serving increasingly sophisticated models requires enormous computing expenditure.
The Fable 5 spending data therefore matters beyond one product launch.
It offers investors a glimpse into whether customers are willing to pay enough for frontier AI to support the enormous capital required to build it.
What is the biggest lesson from Fable 5?
The AI market is discovering that “best” and “most valuable” are not necessarily the same thing.
A company paying for AI is not buying a benchmark score. It is buying an outcome.
If a less expensive model completes the job accurately enough, the economic incentive is obvious.
That could force frontier labs to rethink how they compete. Instead of simply trying to build the smartest model, they may need to build systems that deliver the best result for every dollar spent.
Fable 5’s slow corporate uptake is therefore more than a setback for one Anthropic product. It may be an early signal that the AI industry is moving from an arms race over intelligence toward a harder contest over efficiency.
Key takeaways
- Ramp found that Fable 5 accounted for 11.4% of Anthropic model spending among its measured business users in its first month, despite being the company’s most expensive model.
- Fable 5 represented 6% of tokens purchased from Anthropic in the same period.
- Ramp estimates Fable 5 costs about $10 per million tokens, roughly twice the price of GPT-5.6 Sol.
- OpenAI’s GPT-5.6 Sol accounted for 23% of spending on OpenAI models in the same Ramp comparison.
- Fable 5 was temporarily taken offline in June 2026 after US export controls were imposed, then restored globally on July 1.
- OpenAI cut GPT-5.6 Sol API and credit pricing by more than 20% beginning August 21.
- The broader trend suggests businesses are increasingly optimizing AI spending rather than automatically choosing the most capable available model.
