Anthropic’s strongest AI model may no longer be the obvious choice

Anthropic built Claude Fable 5 to handle its hardest coding, research, and long-running agent tasks. But new business spending data suggest many companies are not automatically choosing the most powerful option. Instead, they are comparing capability with price and routing more work to cheaper models.

Ramp data cited by the Financial Times shows Fable 5 has plateaued at about 11% of model-attributed spending on Anthropic products. That does not mean businesses have abandoned Anthropic. It suggests the premium model is becoming a specialist tool while lower-cost options handle more routine work.

Fable 5 is powerful but expensive

Anthropic launched Fable 5 on June 9 and priced it at $10 per million input tokens and $50 per million output tokens. The company positions the model for ambitious, long-running, complex tasks where advanced capability matters.

Ramp’s August AI Index found Fable 5 accounted for only 6% of Anthropic tokens purchased in its model-level sample and 11.4% of dollars spent. That is a notable gap because the model carries Anthropic’s highest broadly available list price for Claude.

The spending pattern challenges an assumption that customers will always move to the newest frontier model. Businesses appear more willing to reserve the most expensive system for difficult problems and use lower-cost models when the extra capability does not justify the cost.

Opus 5 changed the value equation

Anthropic made that trade-off easier when it released Claude Opus 5 on July 24. The company priced Opus 5 at $5 per million input tokens and $25 per million output tokens, exactly half of Fable 5’s listed API price.

Anthropic describes Opus 5 as a major improvement for long-running agents, coding, and professional work. That means buyers no longer face a simple choice between maximum performance and a large drop in capability when they step down from Fable.

Financial Times reporting based on Ramp data says Opus 5 has already surpassed Fable 5 in business spending. The exact share was not disclosed, but the shift supports the idea that price-performance balance is becoming more important than choosing the highest-priced model by default.

Businesses are splitting work across models

Enterprise AI use is increasingly looking like a portfolio rather than a one-model decision. Companies can send demanding coding or research work to a premium system, then route summaries, classification, drafting, or repetitive tasks to cheaper models.

That approach can lower costs without giving up access to frontier capability. Anthropic itself now offers several price points, including Sonnet 5 at $2 per million input tokens and $10 per million output tokens, well below Fable 5.

Ramp also found more companies using model-serving platforms that offer access to open-source and some Chinese-developed models. However, its July data showed these services were mostly used alongside OpenAI and Anthropic, rather than replacing them across the board.

OpenAI is competing aggressively on price

Anthropic is also facing pressure from OpenAI. Ramp’s July data found that GPT-5.6 Sol accounted for 25% of OpenAI tokens and 23% of model-attributed spending, while Fable 5 generated about 75% as much spending as Sol in the same dataset.

OpenAI then cut GPT-5.6 Sol developer pricing on Aug. 21. The current promotional API price is $4 per million input tokens and $20 per million output tokens, making it substantially cheaper than Fable 5 on published token rates.

That difference gives businesses another reason to compare models task by task. A company may still prefer Fable for certain high-value workloads. Still, the financial case becomes harder to make when a competing model can deliver acceptable results at much lower cost.

Anthropic is still growing rapidly

Weak adoption of one premium model does not mean Anthropic’s business is shrinking. Ramp’s July data showed 43.5% of U.S. businesses in its sample paid for Anthropic products that month.

Reuters reported Anthropic’s annualized revenue run rate exceeded $65 billion by the end of July, up from $47 billion in May and about $9 billion at the end of 2025.

The key question is less about overall Anthropic demand and more about how much spending its highest-priced models can capture as customers compare capability, price, and workload needs across tasks.

Ramp’s numbers have important limits

Ramp’s data offers a useful view of business behavior, but it is not a complete picture of the AI market. The company says its model-level usage sample comes from a token-spend management product and skews somewhat toward technology-focused businesses.

Its figures also cover companies using Ramp, not every U.S. business or every Anthropic customer. Large enterprises may have direct contracts, cloud provider agreements, or negotiated pricing that cannot be fully reflected in published token rates and Ramp’s sample.

Other datasets can show different patterns. Vercel’s AI Gateway data, for example, shows Fable 5 still accounting for a measurable share of spending in production traffic. The safest conclusion is that Ramp identifies a real cost-sensitive trend among its customers, not that demand for Fable has disappeared everywhere.

Frontier AI economics may be changing

For AI labs, the larger issue is whether ever more powerful models can continue to command large price premiums. Training frontier systems requires enormous investment, so companies benefit when customers pay more for their most advanced products.

But buyers care about the cost of completing useful work, not simply benchmarking leadership. If a cheaper model performs well enough for most tasks, companies can save money by using the frontier only when the difference matters.

Fable 5 remains an advanced option for demanding work, while Opus 5, Sonnet 5, and competing models offer lower-cost paths for many jobs. The emerging lesson is simple: being the most capable model may no longer guarantee being the default business choice.

TL;DR:

• Anthropic’s Fable 5 costs $10 per million input tokens and $50 per million output tokens, making it far more expensive than several strong alternatives.
• Ramp found that Fable 5 accounted for 6% of Anthropic tokens and 11.4% of model-attributed spending in its model-level sample.
• Claude Opus 5 costs half as much as Fable 5 on listed API token prices and has reportedly surpassed Fable in business spending.
• OpenAI’s GPT-5.6 Sol now costs $4 per million input tokens and $20 per million output tokens under promotional pricing.
• Anthropic remains a strong enterprise AI provider, but businesses increasingly appear to choose models based on price and task rather than maximum capability alone.

Don’t forget to follow us for more exclusive content.

If you liked this, you might also like:

• Anthropic signals possible identity verification steps for Claude users
• Anthropic says its AI spotted vulnerabilities in high-security government systems
• A US AI restriction has Europe racing to secure advanced Anthropic models

This article was made with AI assistance and human editing.