TL;DR
Anthropic’s Claude Fable 5 is priced at $10 per million input tokens and $50 per million output tokens, double Claude Opus 4.8. Third-party benchmarking cited in the source material shows a 5.7% Intelligence Index gain over Opus 4.8, while public productivity evidence remains limited to one unaudited customer story.
Anthropic’s Claude Fable 5 is facing price-performance scrutiny after a review found the model costs twice as much as Claude Opus 4.8 while delivering a 5.7% higher Intelligence Index score in third-party benchmarking. The finding matters for companies setting 2026 AI budgets, because public evidence for major productivity gains remains limited.
The source review says Fable 5 is priced at $10 per million input tokens and $50 per million output tokens, compared with $5 and $25 for Opus 4.8. It says those prices are confirmed by Anthropic’s launch materials, platform pricing documentation and product page, and were also cited by outlets and AI infrastructure firms including Forbes, TechCrunch, Vellum, Finout and Artificial Analysis.
The review cites Artificial Analysis benchmark data showing Fable 5 at 64.9 on its Intelligence Index, compared with 61.4 for Opus 4.8. It also cites GDPval-AA knowledge-work Elo scores of 1,932 for Fable 5 and 1,890 for Opus 4.8, a 42-point gap. The same review says a full Intelligence Index run would cost about $9,940 on Fable 5, compared with about $4,970 on Opus 4.8.
The source material says Fable 5 remains Anthropic’s highest-scoring public model on aggregators. It also says the model and Mythos 5 share the same underlying weights and pricing, with differences tied to safeguards rather than core model weights.
Budget Pressure for AI Buyers
The report’s central point is economic: Fable 5 is measurably stronger than Opus 4.8 on cited benchmarks, but the published average gain is much smaller than the 2x list-price increase. For companies using large volumes of tokens, that difference can become a material budget issue.
The review says the effective price may be lower than the sticker rate because Fable 5 supports 90% prompt-caching discounts, batch pricing at $5 input and $25 output per million tokens, and a blended rate of $7.70 per million tokens under a 7:2:1 cache-hit-to-input-to-output mix. Even with those discounts, buyers still need to test whether their own workloads match the areas where Fable 5 shows stronger gains.
MLOps for Decision Makers: The First Plain-Language Guide to Machine Learning Operations — Why 95% of AI Projects Fail and How to Make Sure Yours Doesn't (The AI Literacy Series Book 4)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Benchmarks Outpace Productivity Proof
The review separates documented pricing from claimed productivity gains. It says the price is well supported by primary and secondary sources, while the public case for enterprise productivity rests on a narrower evidence base.
The main customer example cited in the source material is a Stripe story from Anthropic’s launch post. Anthropic said Fable 5 completed a codebase-wide migration in a 50-million-line Ruby codebase in one day, work that it said would otherwise have taken a team more than two months by hand. The review treats that as a verified statement made by Anthropic, not as an independently audited productivity study.
The review also cautions that average benchmark gaps may hide stronger task-specific results. It says Fable 5 set records in 5 of 10 sub-benchmarks, meaning the model may be more valuable for some hard workloads than the blended score suggests.
“The economics are fully documented. The productivity story is not.”
— Thorsten Meyer AI review
Leadership Transformation for the AI Era: Designing the Vision, Not Just Directing It. The 6×6 Blueprint for Strategic Trust and Decision … series for enterprise transformation)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Open Questions on Real-World Gains
It is still unclear whether Fable 5’s benchmark gains translate into broad enterprise productivity gains large enough to justify its price premium. The source material says there are no controlled human-baseline productivity studies in the public record for the model.
The Stripe example remains a customer story cited by Anthropic, not an independently audited study. Details such as the migration scope, review process, quality bar, human supervision time and repeatability across other companies are not established in the supplied material.
End-to-End AI Evaluation: Building Effective Metrics, Pipelines, and Monitoring for LLM Systems
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Enterprise Pilots Will Decide Value
The next test is practical adoption. Buyers comparing Fable 5, Opus 4.8 and Sonnet 4.6 will need workload-specific pilots that measure accuracy, latency, token use, caching rates and human time saved.
More independent testing could clarify whether the model’s premium is justified for coding migrations, long-context analysis, agentic workflows or other high-value tasks. Until then, the confirmed story is that Fable 5 costs more and scores higher on cited benchmarks, while the broader productivity case is still developing.
large language model input output tokens
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is the confirmed price of Claude Fable 5?
The source material says Claude Fable 5 costs $10 per million input tokens and $50 per million output tokens. That is double the cited price of Claude Opus 4.8.
How much better is Fable 5 than Opus 4.8 on the cited benchmark?
The review cites Artificial Analysis scores of 64.9 for Fable 5 and 61.4 for Opus 4.8, a 5.7% increase on the Intelligence Index.
Does the review say Fable 5 is a bad model?
No. The review says Fable 5 is Anthropic’s highest-scoring public model on aggregators. Its concern is whether the measured gains justify the higher price for many buyers.
What productivity evidence is available?
The source material says the public human-baseline evidence is limited to one unaudited customer story involving Stripe and a large Ruby codebase migration. It says controlled public studies are not yet available.
What should companies do before buying Fable 5 at scale?
Companies should run workload-specific tests comparing Fable 5 with lower-priced models, including cost per completed task, quality, human review time and caching behavior.
Source: Thorsten Meyer AI
