TL;DR
Anthropic’s Claude Fable 5 is reported to be the company’s most expensive active public model, priced at $10 per million input tokens and $50 per million output tokens. The cited benchmark gap over Opus 4.8 is 5.7% on Artificial Analysis’s Intelligence Index, while public productivity evidence is limited to one unaudited customer story.
Anthropic’s Claude Fable 5 has become the company’s most expensive active public model, priced at $10 per million input tokens and $50 per million output tokens, while the public benchmark evidence cited in the source material shows a narrower 5.7% measured intelligence gain over Opus 4.8, a model listed at half the price.
The pricing cited for Claude Fable 5 is presented as confirmed by Anthropic’s launch post, its platform pricing documentation and its product page, and as corroborated by several secondary sources. The source material says Fable 5 costs $10 per million input tokens and $50 per million output tokens, compared with $5 and $25 for Opus 4.8 and roughly $3 and $15 for Sonnet 4.6.
The benchmark comparison cited in the material comes from Artificial Analysis. It reports an Intelligence Index score of 64.9 for Fable 5 versus 61.4 for Opus 4.8, a 5.7% increase. On the GDPval-AA knowledge-work benchmark, the cited result is 1,932 Elo for Fable 5 versus 1,890 for Opus 4.8, which the source frames as a 2.2% difference.
The source material also says Fable 5 supports a 90% prompt-caching discount, a batch tier at $5/$25, and a blended rate of $7.70 per million tokens under a 7:2:1 cache-hit-to-input-to-output pattern. Those figures reduce the effective cost in some workloads, but the listed price remains double Opus 4.8.
Budget Pressure Meets Thin Proof
The central issue for buyers is the gap between documented pricing and documented productivity gains. The source material says the model’s economics are publicly verifiable, while its productivity case rests on one unaudited customer example, not controlled human-baseline studies.
That matters because enterprises planning 2026 AI budgets may be asked to pay a premium for top-line model performance. If the workload benefits from Fable 5’s strongest benchmark areas, the premium may be easier to justify. If the workload resembles the aggregate benchmark average, the cited evidence suggests buyers may face twice the cost for a smaller measured gain.
The finding does not establish that Fable 5 is poor value in every case. The source material says the model is Anthropic’s highest-scoring public model on aggregators and records stronger results in some sub-benchmarks. The question for customers is whether those gains appear in their own tasks, not only in blended public scores.
AI language model pricing
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How Fable Compares
Fable 5 is described in the source material as Anthropic’s strongest public model and as part of a lineup that includes Opus 4.8 and Sonnet 4.6. The material says only older Opus 4 and Opus 4.1 models were priced higher, at $15 per million input tokens and $75 per million output tokens.
The source also clarifies that Fable 5 and Mythos 5 use the same underlying weights and carry the same price, differing in safeguards. That means price comparisons that group the two together are described as accurate in the source material, rather than a category error.
The productivity example cited by the source comes from Stripe, as quoted in Anthropic’s launch material. It says Fable 5 completed a codebase-wide migration in a 50-million-line Ruby codebase in one day, work that “would otherwise have taken a whole team over two months by hand.” The source stresses that this survived verification as a claim made by Anthropic and repeated by others, not as independently audited productivity proof.
enterprise AI language models
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Productivity Evidence Still Missing
It is not yet clear whether Fable 5’s benchmark advantage translates into broad enterprise productivity gains. The source material says there are no controlled human-baseline productivity studies in the public evidence base it reviewed.
The Stripe migration example remains a vendor-cited customer story, not a public audit. The source material says stronger interpretations of that example, such as treating it as independently established human-work evidence, did not survive verification.
Several details also remain workload-dependent, including whether customers can achieve the cited $7.70 blended token rate, whether their tasks fall into the model’s strongest sub-benchmark categories, and whether the premium produces measurable gains after integration, review and operational costs.
AI benchmark comparison tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Enterprise Buyers Need Workload Tests
The next step for customers is likely controlled evaluation against their own workloads. The source material points toward a practical test: compare Fable 5 with Opus 4.8 and lower-cost models on the specific tasks that drive spending, using real prompts, review standards and token patterns.
More public evidence could come from independent benchmarkers, audited customer case studies or Anthropic’s own additional disclosures. Until then, the most firmly established facts are the price, the cited benchmark scores and the absence of broad public proof that the premium consistently produces matching productivity gains.
cost-effective AI models for business
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What happened with Claude Fable 5 pricing?
Claude Fable 5 is listed in the source material at $10 per million input tokens and $50 per million output tokens, making it twice the price of Opus 4.8 on the cited rates.
Does Fable 5 perform better than Opus 4.8?
According to the cited Artificial Analysis data, yes, but the aggregate gap is smaller than the price gap: 64.9 versus 61.4 on the Intelligence Index, or 5.7%.
Is there proof that Fable 5 makes workers more productive?
The source material says the public evidence base contains one unaudited customer story and zero controlled human-baseline productivity studies. That means productivity gains remain a claim to test, not a settled finding.
Could the higher price still be justified?
Yes, for some workloads. The source material says Fable 5 performs better in certain sub-benchmarks, so the case depends on whether a buyer’s actual tasks match the areas where the model’s gains are strongest.
What should enterprise buyers do next?
Buyers should run side-by-side workload tests against Fable 5, Opus 4.8 and cheaper alternatives, measuring output quality, review time, token cost and deployment fit before committing budget.
Source: Thorsten Meyer AI
