🔍 Read the full analysis: AI Intelligence Rises 5.7% As The Price Doubles on ThorstenMeyerAI.com
Get smart everyday buys 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 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 scores Fable 5 5.7% higher on its Intelligence Index, but the available source material identifies no controlled human-baseline productivity study and one unaudited customer testimonial.
Anthropic’s Claude Fable 5 costs twice as much per token as Claude Opus 4.8, while scoring 5.7% higher on Artificial Analysis’s Intelligence Index, according to a July 2026 analysis of the model’s pricing and publicly available performance evidence. The analysis says the case for higher workplace productivity rests publicly on one unaudited customer account, rather than a controlled study measuring output against a human or no-AI baseline.
Anthropic lists Fable 5 at $10 per million input tokens and $50 per million output tokens. The corresponding listed prices for Opus 4.8 are $5 and $25, while Sonnet 4.6 is listed at about $3 and $15. That makes Fable 5 the most expensive model in Anthropic’s active lineup, at twice Opus 4.8’s price and roughly 3.3 times Sonnet 4.6’s, based on the figures in the source analysis.
The performance comparison is narrower than a claim that Fable 5 is broadly 5.7% better at every task. Artificial Analysis gives it an Intelligence Index score of 64.9, compared with 61.4 for Opus 4.8. On the separate GDPval-AA knowledge-work benchmark, the scores are 1,932 and 1,890, respectively. The analysis also says Fable 5 sets records in five of 10 sub-benchmarks, suggesting results vary by task.
The analysis reports a blended rate of $7.70 per million tokens under an assumed mix of seven cache-hit tokens, two input tokens and one output token. That is a usage scenario, not the model’s list price: actual costs depend on the balance of cached input, uncached input and output. The source also cites discounted batch pricing and a higher rate for US-only inference.
The Cost of a Small Average Gain
For businesses choosing an AI model, a benchmark lead does not by itself establish that the higher-priced option will save enough staff time or money to justify its cost. Fable 5’s 2× price relative to Opus 4.8 is directly comparable on the listed token rates; the practical value depends on the task, usage pattern and whether the model’s advantages translate into better results for a particular organization.
The benchmark figures also need to be read in their proper scope. A 5.7% Intelligence Index difference is an aggregate score, not a measured productivity increase. The separate GDPval-AA result is a 42-point Elo difference; the source analysis cautions that turning this rating into a simple percentage can obscure how the scale works. Neither result establishes how much faster employees complete work, how much human review remains necessary or whether the added cost is offset by gains.
That distinction matters to budget decisions. The source presents pricing as confirmed by Anthropic’s own materials and benchmark results as third-party measurements, but says it found no controlled Fable 5 study against a human baseline. Organizations should treat the productivity case as less established than the model’s price and benchmark scores.
AI language model API subscription
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Pricing, Benchmarks and the Launch Claim
Fable 5 launched on June 9, 2026, with general availability listed around July 1 in the source material. The analysis says Anthropic’s launch post, platform pricing documentation and product page list the model’s rates, which it says are also reported by several technology and industry publications. The evidence for price is therefore distinct from evidence about workplace outcomes.
The central customer example in Anthropic’s launch post concerns a 50-million-line Ruby codebase. Anthropic attributed to Stripe the statement that Fable 5 completed a codebase-wide migration in one day, a task that would otherwise have taken a team more than two months by hand. The analysis confirms that the statement appeared in the launch post; it does not establish the result through an independent audit or replication.
The source analysis says stronger descriptions of that example as an independently established human-work benchmark were not supported by its verification process. It characterizes the case as a vendor-published customer testimonial, without published methodology. That makes the story relevant as an illustration of a reported use, but not equivalent to a controlled productivity study.
“The migration would otherwise have taken a whole team over two months by hand.”
— Anthropic’s launch post, attributing the account to Stripe
AI model performance benchmarking tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What the Productivity Evidence Cannot Show
The material provided does not identify a controlled productivity study comparing Fable 5 with human-only work or a no-AI baseline. It also does not provide an independent audit or replication of the Stripe account. The time-saving claim should therefore remain attributed to the customer account as presented by Anthropic, rather than treated as a general performance result.
It is also unclear how the benchmark gains would translate to any particular company’s tasks, costs or quality requirements. The overall index combines performance across evaluations, while the source says results differ across sub-benchmarks. The material does not establish which workloads will benefit enough to justify the premium, or whether the reported customer example is typical.
The source’s comparison with GPT-5.5 is described as resting on one secondary source, so it is directional rather than firmly established. The headline comparison with Opus 4.8 is more directly supported by the listed prices and benchmark figures. Neither comparison is a guarantee of real-world cost or productivity for a specific user.
As an affiliate, we earn on qualifying purchases.
What Buyers Should Verify Next
The next useful evidence would be independent, repeatable evaluations of Fable 5 on defined workplace tasks, with clear information on completion time, quality, human review and total token costs. Studies comparing the model with both Opus 4.8 and human-only workflows would help separate the value of the model’s benchmark lead from the value of using AI at all.
Until that evidence is available, organizations weighing the upgrade can test Fable 5 on their own representative workloads and compare results against a lower-priced model. Such trials can measure task-specific performance and actual usage costs, but a company’s internal test would not by itself establish a general productivity claim. The source material does not give a date for further independent studies or additional audited customer results.
As an affiliate, we earn on qualifying purchases.
Key Questions
How much does Claude Fable 5 cost?
The source lists Anthropic’s standard prices as $10 per million input tokens and $50 per million output tokens. Usage costs can vary with caching, batch pricing and the share of input versus output tokens.
What does the 5.7% figure measure?
It is the difference between Fable 5’s 64.9 Intelligence Index score and Opus 4.8’s 61.4 score on Artificial Analysis’s index. It is not a direct measurement of employee productivity or time saved.
Is there a controlled study showing Fable 5 increases productivity?
The source analysis says it found no controlled human-baseline productivity study for Fable 5. It identifies one customer account in Anthropic’s launch post, attributed to Stripe, and says that account was not independently audited or replicated in the material reviewed.
Does Fable 5 outperform Opus 4.8 on every task?
The source does not establish that. It reports higher aggregate scores for Fable 5 and records in five of 10 sub-benchmarks, meaning results differ across the evaluated tasks.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
