🔍 Read the full analysis: The AI Upgrade Math: Twice The Price, 5.7% More Intelligence on ThorstenMeyerAI.com
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
TL;DR
Anthropic’s Claude Fable 5 costs $10 per million input tokens and $50 per million output tokens, twice Opus 4.8’s listed rates. Artificial Analysis reports a 5.7% higher Intelligence Index score. The source analysis identifies one unaudited customer account and no controlled productivity studies.
Anthropic’s Claude Fable 5 is priced at $10 per million input tokens and $50 per million output tokens, twice the listed rates for Opus 4.8. Artificial Analysis reports a 5.7% higher Intelligence Index score for Fable 5. A July analysis by ThorstenMeyerAI.com says the public evidence it reviewed on enterprise productivity consists of one unaudited customer account, with no controlled studies comparing results against a human-only baseline.
The price figures are documented in Anthropic’s launch post and pricing materials, according to the analysis, and are corroborated by several secondary publications and model-information services. Opus 4.8 is listed at $5 per million input tokens and $25 per million output tokens; Sonnet 4.6 is listed at about $3 and $15, respectively. Fable 5’s listed rates are about 3.3 times Sonnet 4.6’s.
On the Intelligence Index cited in the report, Artificial Analysis scores Fable 5 at 64.9, compared with 61.4 for Opus 4.8. The analysis calculates the difference as 5.7%. On GDPval-AA, a knowledge-work benchmark expressed as Elo, Fable 5 scores 1,932 against 1,890. These results measure benchmark performance, not how much more work employees complete in a company.
The report identifies one public customer example: Anthropic said a Fable 5 project migrated a 50-million-line Ruby codebase in a day, a task the company said would otherwise have taken a team more than two months by hand. The claim is attributed to Stripe in Anthropic’s launch post. ThorstenMeyerAI.com says it found no published methodology, independent audit or replication for the account, and no controlled enterprise productivity study comparing Fable 5 with a human-only baseline.
Comparing Price and Benchmark Results
Companies evaluating models can compare Fable 5’s listed token price with its benchmark results, but the benchmark advantage does not by itself establish lower costs, faster delivery or greater output across a company’s work.
The source analysis says Fable 5 leads in five of ten sub-benchmarks, with gains varying across tasks. Whether those results apply to a particular organization depends on how closely its work resembles the benchmark tasks and whether performance differences justify paying twice Opus 4.8’s listed rate.
Token prices are not a complete operating-cost comparison. The analysis describes discounts for cached prompts and batch processing, as well as higher rates for US-only inference. Actual costs depend on usage patterns, output volume and configuration. Buyers can evaluate representative tasks by tracking quality, review time and total cost alongside benchmark results.
As an affiliate, we earn on qualifying purchases.
Launch Pricing and Benchmark Comparisons
Anthropic launched Fable 5 on June 9, 2026, according to the source analysis, which listed general availability as expected around July 1. The report describes it as Anthropic’s most expensive model in its active lineup at the time. It presents pricing and model comparisons separately from productivity claims.
The listed rates do not cover every billing scenario. The analysis gives a $1 per million-token cache-read rate and batch pricing of $5 per million input tokens and $25 per million output tokens. It estimates a blended price of $7.70 per million tokens using a stated 7:2:1 ratio of cache-hit, input and output tokens. This is an illustrative calculation based on that mix, not a universal rate. US-only inference is listed at a 1.1-times multiplier.
The analysis also says Fable 5 and Mythos 5 share the same underlying weights and are priced identically, with safeguards distinguishing the products. It cautions that the 42-point GDPval-AA Elo difference is not a percentage: Elo is a rating scale, while the Intelligence Index is a separate composite measure. Neither result is a controlled workplace productivity trial.
“would otherwise have taken a whole team over two months by hand”
— Anthropic’s launch post, quoting Stripe
As an affiliate, we earn on qualifying purchases.
Public Productivity Evidence
The available material does not establish how Fable 5 performs against a human-only or no-AI baseline in a controlled enterprise study. The source analysis reports no such controlled studies and identifies one unaudited customer testimonial. This describes the public evidence reviewed by the analysis; it does not establish whether private customer evaluations exist.
The report does not establish whether the Stripe migration is representative of other organizations, how much human review or preparation it required, or whether the comparison accounts for testing, correction and deployment. The source does not provide independently verified measurements for those factors. Whether the benchmark results translate into savings or increased output for a particular customer remains unconfirmed.
Benchmark scores and prices can change, and the supplied material does not establish how performance applies across all tasks. The report’s account of five sub-benchmark leads does not provide enough detail here to identify which buyer workloads may benefit most. Comparisons with other companies’ models are also less certain when they rely on secondary sources rather than current, comparable primary documentation.
As an affiliate, we earn on qualifying purchases.
Evaluate Models on Representative Tasks
Further evidence could come from controlled evaluations on representative workloads that document the model, prompts and comparison method. Companies can compare Fable 5 with lower-priced options using the same tasks and track completion time, error rates, human review and total token costs. These measurements can show how the models perform in specific work settings.
Additional public customer studies could describe the Stripe migration’s scope, human involvement, quality checks and the method used to calculate the time comparison. Independent replication could help assess whether the result applies beyond that project. The available public material documents Fable 5’s listed prices and benchmark scores but does not establish a general productivity return.
As an affiliate, we earn on qualifying purchases.
Key Questions
How much does Claude Fable 5 cost?
The source analysis lists $10 per million input tokens and $50 per million output tokens. It also reports lower rates for batch processing and cache reads, plus a multiplier for US-only inference.
What does the 5.7% figure measure?
It is the difference reported by the analysis between Fable 5’s 64.9 Intelligence Index score and Opus 4.8’s 61.4 score on Artificial Analysis. It is not a measured 5.7% increase in workplace productivity.
What public evidence supports Fable 5 productivity claims?
The analysis identifies one customer account in Anthropic’s launch post, attributed to Stripe, about migrating a 50-million-line Ruby codebase in a day. It says the account is unaudited and lacks public methodology or independent replication, and reports no controlled human-baseline studies.
Does the benchmark lead mean Fable 5 is worth twice as much?
Not by itself. The benchmark measures model performance on defined evaluations, while value depends on the buyer’s tasks, usage and total costs. The source material does not establish a general productivity return that offsets the higher list price.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
