🔍 Read the full analysis: Breaking Down The Cost Of 5.7% More AI Intelligence on ThorstenMeyerAI.com
Get hardware and tech essentials delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Anthropic’s Fable 5 is priced at $10 per million input tokens and $50 per million output tokens, twice Opus 4.8’s rates. Artificial Analysis reports a 5.7% higher Intelligence Index score, but the cited enterprise productivity evidence is one unaudited customer account, not a controlled study.
Anthropic’s Claude Fable 5 costs twice as much per token as Opus 4.8, while a third-party benchmark reports a 5.7% higher Intelligence Index score—a gap that leaves buyers with firmer evidence on price than on workplace productivity. A July 2026 analysis by Thorsten Meyer of ThorstenMeyerAI.com says the publicly cited enterprise productivity case amounts to one unaudited customer account, not a controlled study.
Anthropic lists Fable 5 at $10 per million input tokens and $50 per million output tokens, according to its launch post and pricing materials cited by Meyer. Opus 4.8 is listed at $5 and $25, respectively. Sonnet 4.6 is listed at about $3 and $15. That makes Fable 5 the most expensive model in Anthropic’s active lineup, based on the source analysis.
Artificial Analysis scores Fable 5 at 64.9 on its Intelligence Index, compared with 61.4 for Opus 4.8. The source analysis describes the difference as 5.7%. On the GDPval-AA knowledge-work benchmark, Fable 5 scores 1,932 Elo to Opus 4.8’s 1,890, a reported 42-point gap. The analysis cautions that a percentage comparison of Elo scores does not fully express what the rating difference means in head-to-head terms.
The analysis also reports that a full Intelligence Index run costs about $9,940 for Fable 5, versus about $4,970 for Opus 4.8. Those figures describe benchmark evaluation costs, not the price of a typical customer task. Actual bills vary with token use, caching, batch processing and workload design.
Model economics · July 2026 analysis
Breaking Down the Cost of 5.7% More AI Intelligence
Claude Fable 5 is listed at twice Opus 4.8’s token prices, while a third-party benchmark reports a 5.7% higher Intelligence Index score. Public evidence for workplace productivity remains much thinner than the price data.
Price is the clearest comparison
Rates below are per million tokens, as reported in Anthropic’s launch and pricing materials cited by the analysis. Actual bills vary with token use, caching, batch processing and workload design.
Claude Fable 5
Highest listedThe most expensive model in Anthropic’s active lineup, according to the source analysis.
Claude Opus 4.8
ReferenceFable 5’s listed input and output prices are both double these rates.
Claude Sonnet 4.6
Lower costA lower-priced active-lineup point of comparison in the cited analysis.
A measurable lead, with limits
Artificial Analysis reports higher scores for Fable 5 on two measures cited by the source analysis. The results do not directly measure productivity in a company’s operations.
Intelligence Index
Reported index scores · scale shown from zero to 70
GDPval-AA knowledge work
Reported Elo ratings · close-up scale from 1,800 to 1,950
Performance on benchmark tasks under their defined evaluation conditions.
A general workplace productivity percentage, payback period or typical customer savings.
The productivity case is unverified
The July 2026 analysis separates published rates and benchmark results from reported customer outcomes. It identifies one unaudited enterprise account, not a controlled study.
Published rates
Anthropic launch and pricing materials support the listed token rates. These are concrete prices, though actual usage costs depend on workload and applicable rates.
Comparative scores
Artificial Analysis provides the cited Intelligence Index and GDPval-AA comparisons. Scores are specific measures, not direct evidence of business output.
Customer productivity
No public methodology, audit or independent replication is cited for the enterprise productivity account. It cannot show what other teams should expect.
“The economics are fully documented. The productivity story is not.”Thorsten Meyer · ThorstenMeyerAI.com
Anthropic’s launch post says Fable 5 completed a migration across a 50-million-line Ruby codebase in one day, a job described as taking a team more than two months by hand. The analysis says the account lacks public detail on methodology, quality checks, staffing assumptions and full costs. Repetition of the claim by other outlets does not independently verify it.
Make the premium prove itself
The right choice depends on task mix and the value of any improvement. Companies should measure against their own work before projecting savings at scale.
Choose real tasks
Use representative work from the organization, including difficult cases and routine volume.
Set a baseline
Compare with a human process or existing model using the same inputs and success criteria.
Measure quality
Track completion, errors, review effort and time—not speed alone.
Count total cost
Include input and output tokens, caching, batch rates, human review and failures.
Key questions
A benchmark lead is not the same as a measured business gain.
How much does Claude Fable 5 cost?
The cited analysis reports $10 per million input tokens and $50 per million output tokens. Actual costs depend on usage and applicable caching or batch pricing.
How does it compare with Opus 4.8?
Fable 5’s listed input and output rates are twice Opus 4.8’s. Artificial Analysis reports Intelligence Index scores of 64.9 and 61.4, respectively.
Does 5.7% more benchmark score mean 5.7% more productivity?
No. The figure refers to the Intelligence Index comparison, not a measured gain in employee output. The reviewed public evidence contains no controlled human-baseline productivity study.
What is the Stripe productivity claim?
Anthropic’s launch post says Fable 5 completed a migration across a 50-million-line Ruby codebase in one day. The cited analysis says the public account is unaudited and lacks enough detail to establish a general result.
For companies choosing a model for large-scale use, the comparison is a question of marginal performance against added cost. Fable 5’s headline price is twice Opus 4.8’s, while its overall Intelligence Index lead is narrower. That does not establish that Fable 5 is poor value: the benchmark is an aggregate, and results can differ by task. But it does mean buyers cannot assume a higher benchmark score will translate into a proportionate productivity gain across their organization.
Meyer reports that Fable 5 sets records in five of ten sub-benchmarks. The source argues that improvements on particular difficult tasks may be more meaningful to some buyers than the overall average. Companies whose workloads resemble those tasks could reach a different cost-benefit judgment from companies running routine work. The public material cited does not establish which organizations will see enough additional output to offset the premium.
The evidence gap matters to budget planning. A vendor price is relatively straightforward to verify; a productivity claim depends on a defined baseline, comparable work, and measurement of quality as well as speed. Without those details, organizations should treat the cited customer story as an example of reported use, not a general estimate of expected savings. This is a report on model economics, not financial advice; AI costs and outcomes vary, and expenditures may not produce a return.
AI language model pricing comparison
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What the Public Evidence Shows
The analysis separates documented pricing and benchmark results from claims about productivity. It says Anthropic’s launch post, platform pricing documents and product page support Fable 5’s listed rates. It also cites Artificial Analysis as the source for the Intelligence Index and GDPval-AA comparisons. Those figures offer specific, comparable measures, but they do not directly measure productivity in a company’s own operations.
For the human-baseline productivity case, Meyer identifies a statement in Anthropic’s launch post about Stripe. According to that account, Fable 5 completed a migration across a 50-million-line Ruby codebase in one day, a job described as otherwise taking a team more than two months by hand. The analysis says this statement was repeated by other outlets, but repetition does not independently verify the result or its comparison with manual work.
Meyer says stronger descriptions of the Stripe example as an independently established baseline did not withstand the analysis’s verification process. The cited account has no published methodology, audit or independent replication in the source material. The distinction is important: the account confirms what Anthropic said about a customer result, but does not prove that other teams will achieve similar gains.
“The economics are fully documented. The productivity story is not.”
— Thorsten Meyer, ThorstenMeyerAI.com
As an affiliate, we earn on qualifying purchases.
Productivity Gains Remain Unverified
The source analysis says it found no controlled human-baseline productivity study for Fable 5 and one unaudited enterprise proof point. The available material does not specify a public, independent assessment of the Stripe account’s methodology, quality checks, staffing assumptions or full costs. It is also not clear whether the reported one-day migration included substantial human planning, review or correction.
Benchmark scores cannot answer those questions on their own. They measure performance under defined evaluation conditions, while business results depend on a team’s tasks, tools, review requirements and error tolerance. The source does not provide enough evidence to calculate a general productivity percentage, a payback period, or the likely savings for a typical customer. Its summary of the evidence base is the analysis author’s finding, rather than a complete audit of every possible private customer study.
As an affiliate, we earn on qualifying purchases.
Measure Results on Real Workloads
The source material does not identify a scheduled independent productivity study or a forthcoming change to Fable 5’s pricing. For now, the next useful evidence would be controlled tests on representative work, with a clear human or existing-model baseline, quality measures, time accounting and total usage costs. Independent replication of customer examples would also help establish whether reported results extend beyond a single deployment.
Organizations weighing the model can compare its performance and cost on their own tasks against Opus 4.8 and lower-priced alternatives. Such evaluations should account for input and output tokens, cache use, batch rates, human review and failures, rather than relying only on list prices or aggregate benchmark rankings. Until broader evidence is published, the productivity case remains less settled than the price comparison.
AI productivity benchmarking software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
How much does Claude Fable 5 cost?
The source analysis reports Anthropic list prices of $10 per million input tokens and $50 per million output tokens. Actual costs depend on usage and applicable caching or batch pricing.
How does Fable 5 compare with Opus 4.8?
Fable 5’s listed input and output rates are twice Opus 4.8’s. Artificial Analysis reports an Intelligence Index score of 64.9 for Fable 5 and 61.4 for Opus 4.8.
Does the 5.7% benchmark lead mean 5.7% more workplace productivity?
No. The 5.7% figure refers to the Intelligence Index comparison, not a measured gain in employee output. The source analysis says there is no controlled human-baseline productivity study for Fable 5 in the public evidence it reviewed.
What is the Stripe productivity claim?
Anthropic’s launch post says Fable 5 completed a migration in a 50-million-line Ruby codebase in one day, compared with work the company said would otherwise have taken a team more than two months by hand. The analysis describes this as an unaudited customer account, not an independently verified study.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
