Read the full analysis: Twice The Cost For A Little More AI Intelligence on ThorstenMeyerAI.com
TL;DR
Anthropic’s Claude Fable 5 is priced at $10 per million input tokens and $50 per million output tokens, twice the rates for Opus 4.8. Artificial Analysis reports a 5.7% higher Intelligence Index score, while the public human-baseline productivity evidence cited in the source consists of one unaudited Stripe account.
Anthropic’s Claude Fable 5 costs $10 per million input tokens and $50 per million output tokens, twice the listed rates for Opus 4.8, while scoring 5.7% higher on Artificial Analysis’s Intelligence Index. A July 2026 report from ThorstenMeyerAI.com says the public case for greater enterprise productivity is much thinner: it identifies one unaudited customer anecdote and no controlled human-baseline productivity study for Fable 5.
The pricing figures are presented in the report as confirmed by Anthropic’s launch post and pricing materials. 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, making Fable 5 roughly 3.3 times as expensive as Sonnet on those rates. The report describes Fable 5 as the most expensive model in Anthropic’s active lineup.
On Artificial Analysis’s Intelligence Index, Fable 5 scores 64.9, compared with 61.4 for Opus 4.8, a reported 5.7% difference. On GDPval-AA, a knowledge-work benchmark measured in Elo points, the scores are 1,932 and 1,890. The report cautions that turning that 42-point Elo gap into a percentage can obscure how the rating scale works. It also says Fable 5 set records in five of the ten sub-benchmarks, so the overall average may not reflect results on every task.
The productivity example cited is Anthropic’s account of work in a 50-million-line Ruby codebase. The company said Fable 5 completed a codebase-wide migration in a day, a task that would otherwise have taken a team more than two months by hand, according to Stripe. ThorstenMeyerAI.com found no published methodology, independent audit or replication for that account. It therefore treats the example as a customer testimonial, not a controlled measurement of typical productivity.
AI Economics · July 2026 Briefing
Twice The Cost For A Little More AI Intelligence
Claude Fable 5 doubles Opus 4.8’s listed token rates for a 5.7% lead on one intelligence index. Public evidence for broader workplace productivity remains a single unaudited customer account.
The customer claim
“The migration would otherwise have taken a whole team over two months by hand.”
Anthropic’s launch post, describing a Stripe migration in a 50-million-line Ruby codebase. No published audit or independent replication was identified.
01 / What the list price buys
A clear premium at checkout
The listed rates are presented in the source report as confirmed by Anthropic’s launch post and pricing materials. Actual spend varies with usage, caching, batch processing, and output volume.
Highest listed rate
Claude Fable 5
The report describes Fable 5 as the most expensive model in Anthropic’s active lineup.
Reference model
Claude Opus 4.8
Fable 5 is twice these listed rates for both input and output.
Lower-cost comparison
Claude Sonnet 4.6
At those rates, Fable 5 is roughly 3.3× as expensive.
02 / Benchmarks
A modest index lead, with task-level variation
Third-party benchmark scores compare model performance on defined tests. They do not directly measure paid work completed by employees.
Artificial Analysis Intelligence Index
Reported composite scores; bars share a 0–70 scale.
Relative difference reported for the Intelligence Index.
GDPval-AA knowledge-work benchmark
Reported Elo ratings; the scale is not a percentage measure.
Turning an Elo gap into a percentage can obscure how the rating scale works.
The report says Fable 5 placed first in 5 of 10 Intelligence Index sub-benchmarks; the overall average may not describe every task.
03 / Price mechanics
Effective cost depends on the mix
List price is one input to a budget. Cache reads, batch rates, and inference location can change the bill, while workload patterns determine how much those options matter.
Pricing detailReported figureHow to read it
Standard cache read$1 / million tokensLower rate for cached input reads
Illustrative blended mix$7.70 / million tokensCalculated with a 7:2:1 mix of cached reads, ordinary input, and output
Batch pricing$5 input · $25 outputReported per million tokens
US-only inference1.1× multiplierReported premium relative to standard pricing
The blended rate is an example, not a universal price. The report also says a full Intelligence Index run costs about $9,940 for Fable 5 versus $4,970 for Opus 4.8.
04 / What the evidence supports
Benchmark results are not a productivity study
The source distinguishes documented pricing, third-party benchmark results, and a customer story. Each supports a different kind of conclusion.
Documented comparison
Rates and benchmark scores
Pricing figures are attributed to Anthropic’s published materials. Artificial Analysis offers an external benchmark comparison, not a measure of typical company output.
Useful for comparing published rates and test scores
Does not show time saved across organizations
Limited public evidence
One Stripe migration account
Anthropic said Fable 5 completed a codebase-wide migration in a day. The report found no published methodology, independent audit, replication, or controlled human-baseline study.
Treat as an attributed customer testimonial
Representativeness, review effort, and time saved are unknown
05 / A practical buying test
Make the premium earn its place
Fable 5 may suit work that matches its stronger benchmark areas. The report says it leads on five of ten sub-benchmarks, but does not establish how often those tasks match real customer workloads. Measure your own tasks before forecasting a return.
Quick answers
What buyers should know
The report’s conclusion is conditional: price and benchmark figures are available, while general productivity gains remain unverified beyond one account.
How much does Fable 5 cost?
$10 per million input tokens and $50 per million output tokens at listed rates, twice the reported Opus 4.8 prices.
How much higher did it score?
Artificial Analysis reports an Intelligence Index score of 64.9 versus 61.4, a 5.7% difference. GDPval-AA scores are 1,932 and 1,890 Elo.
Is workplace productivity proven?
The source report found no controlled human-baseline productivity study and identifies one unaudited account attributed to Stripe.
Could the premium be worthwhile?
Possibly, for a workload that benefits from Fable 5’s strengths. Buyers need to test their tasks and include usage and human review costs.
The Premium Depends on the Workload
The comparison matters to organizations deciding whether a higher-priced model merits a place in an AI budget. At list price, Fable 5 costs twice as much as Opus 4.8. The benchmark results show an average lead on one index, but they do not by themselves establish that a particular company will complete work faster, reduce staffing costs or produce more output.
That distinction leaves the buying decision dependent on the task. The report says Fable 5 leads on five of ten sub-benchmarks, with larger gains on some individual tasks. For a workload that closely matches those areas, the price premium might be worthwhile; for other work, the aggregate scores offer no assurance of a comparable benefit. Buyers need to test their own use cases and account for token usage, caching and output volume rather than treating a benchmark score as a productivity forecast.
The evidence gap also affects how companies should interpret launch claims. Pricing is directly documented, while the available productivity example is a single account attributed to a customer. That does not show the anecdote is false; it means readers cannot infer from it alone how the model performs across teams or organizations.
Pricing, Benchmarks and Launch Claims
The report distinguishes the listed rate from a lower effective cost when prompt caching is used. Anthropic’s standard cache-read rate is given as $1 per million tokens, and the report calculates a blended rate of $7.70 per million tokens using a 7:2:1 ratio of cached reads, ordinary input and output. That is an illustrative usage mix, not a universal price. It also lists batch pricing at $5 per million input tokens and $25 per million output tokens, and a 1.1-times multiplier for US-only inference.
Artificial Analysis’s figures provide a third-party benchmark comparison, not a direct measure of paid work completed by employees. The report says the full Intelligence Index run costs about $9,940 for Fable 5 versus $4,970 for Opus 4.8. It also notes that Fable 5 and Mythos 5 share underlying weights and have the same listed price, with safeguards distinguishing the products. The report dates Fable 5’s launch to June 9, 2026, and says general availability was expected around July 1.
“The migration would otherwise have taken a whole team over two months by hand.”
— Anthropic, as quoted in its launch post and summarized by ThorstenMeyerAI.com
How Much Productivity Is Proven?
The material does not establish how representative the Stripe migration was, what methods were used to measure the time saved, or how much human review was required. It also does not identify a controlled study comparing Fable 5-assisted work with a human or no-AI baseline. The source report says the public evidence it found contains one unaudited enterprise account, not independent proof of a general productivity gain.
Benchmark performance also cannot settle whether a buyer will recover the price premium. The report provides index and Elo scores but no workload-specific cost analysis for a typical organization. Results may differ by task, prompting, caching, output length and review requirements. The source does not provide enough information to determine how frequently the five stronger sub-benchmarks match real customer workloads.
Test Costs on Real Tasks
Organizations considering Fable 5 can compare it with Opus 4.8 on representative tasks, tracking both token costs and completed work. A useful evaluation would define the baseline in advance, include human review time, and measure task quality as well as speed. That would help determine whether any gains on a specific workload offset the higher listed rates.
The source report provides no schedule for an independent productivity study or further audited customer data. Until such evidence is published, the confirmed basis for comparison remains Anthropic’s pricing, third-party benchmark results and the attributed Stripe example. Buyers should treat broader productivity claims as unverified beyond that account rather than as a demonstrated average outcome.
Key Questions
How much does Claude Fable 5 cost?
The source report lists $10 per million input tokens and $50 per million output tokens. It says those rates are twice the listed prices for Opus 4.8.
How much higher did Fable 5 score than Opus 4.8?
Artificial Analysis’s Intelligence Index score is reported as 64.9 for Fable 5 and 61.4 for Opus 4.8, a 5.7% difference. On GDPval-AA, the reported scores are 1,932 and 1,890 Elo.
Is there controlled evidence that Fable 5 boosts workplace productivity?
The source report says it found no controlled human-baseline productivity study for Fable 5. It identifies one Anthropic launch-post account about work attributed to Stripe, without a published audit or independent replication.
Could Fable 5 still be worth the higher price?
Possibly, depending on the work. The report says Fable 5 recorded the top result in five of ten sub-benchmarks, but benchmark scores do not establish results for any particular company. Buyers would need to test their own tasks and include usage and review costs.
Source: ThorstenMeyerAI.com