🔍 Read the full analysis: A Better AI By 5.7%—At Twice The Price on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get the latest gadgets delivered free with Prime

• Fast, free delivery on millions of items
• Prime Video, Amazon Music and more included
• Member-only deals all year

Start your free Prime trial →Free trial for eligible customers · Cancel anytime

As an affiliate, we earn on qualifying purchases.Hide

Anthropic’s Claude Fable 5 costs twice as much as Opus 4.8, while a third-party benchmark puts its Intelligence Index score 5.7% higher; the public evidence for greater enterprise productivity, however, rests on one unaudited customer anecdote. The comparison matters to companies weighing whether the model’s additional capability justifies its higher per-token cost.

Anthropic lists Fable 5 at $10 per million input tokens and $50 per million output tokens. Opus 4.8 is listed at $5 and $25, respectively, while Sonnet 4.6 is listed at approximately $3 and $15. Those prices make Fable 5 the most expensive model in Anthropic’s active lineup, according to the source report, which says the figures are supported by Anthropic pricing materials.

Artificial Analysis reports an Intelligence Index score of 64.9 for Fable 5, compared with 61.4 for Opus 4.8. The same source gives Fable 5 a GDPval-AA knowledge-work Elo rating of 1,932, against Opus 4.8’s 1,890. The report describes the respective differences as 5.7% on the index and 2.2% on GDPval-AA; these are benchmark results, not direct measurements of workplace productivity.

The productivity claim most often cited is a statement in Anthropic’s launch post about a Stripe project. Anthropic says Fable 5 completed a migration across a 50-million-line Ruby codebase in a day, a task that would otherwise have taken a team more than two months by hand. The source report says this remains a single customer testimonial: it found no methodology, independent audit, or replication establishing the comparison as a measured productivity result.

The Cost of a Small Average Gain

For buyers, the central issue is not simply whether Fable 5 scores higher. It is whether the extra capability appears on the workloads a business actually runs, and whether those gains offset twice the usage price of Opus 4.8. A benchmark average can inform that decision, but cannot establish the return on investment for a particular team or task.

The source report estimates that a full Intelligence Index run costs about $9,940 for Fable 5, versus $4,970 for Opus 4.8. That estimate illustrates the cost difference under the benchmark’s usage assumptions; it is not a forecast of a company’s bill. Actual spending depends on token volumes, caching, batching, and the mix of input and output tokens.

The report also cautions against treating the average score as a verdict for every use. It says Fable 5 set records in five of ten sub-benchmarks, suggesting that some task-specific gains may be larger than the aggregate figure. That possibility could matter to teams with demanding workloads, but companies would need workload-specific testing to establish whether those gains apply to them.

How Pricing Meets Benchmark Scores

The comparison is between Fable 5 and Opus 4.8, the Anthropic model priced at half as much. The figures in the source report distinguish list prices from a blended estimate: using a stated mix of seven cache-hit tokens for every two uncached input tokens and one output token, it calculates a blended rate of $7.70 per million tokens for Fable 5. The result depends on that assumed mix, rather than representing a universal rate.

The report also describes prompt caching as eligible for a 90% discount, with cache reads priced at $1 per million tokens, and gives batch pricing of $5 per million input tokens and $25 per million output tokens. It says US-only inference carries a 1.1× multiplier. These options can affect costs, but do not change the headline list-price comparison.

Fable 5 launched on June 9, 2026, and the source places general availability around July 1. It says Fable 5 and Mythos 5 share underlying model weights and have identical pricing, with safeguards as the stated difference. The report’s claims about benchmark scores and pricing should be read in their respective contexts: vendor pricing materials establish prices, while third-party benchmarks measure performance on defined tests.

“would otherwise have taken a whole team over two months by hand”

— Anthropic’s launch post, describing a Stripe project

What the Productivity Claim Establishes

The Stripe example does not, on the evidence described in the source report, establish a general productivity gain for Fable 5. Details such as how the work was scoped, how quality was assessed, how much human review was required, and how the comparison timeline was determined are not provided in the account. The report says it found no controlled human-baseline enterprise study or independent replication.

Benchmark results also leave open how well the scores predict performance on a buyer’s own work. The source gives aggregate and sub-benchmark results, but does not establish that a particular organization’s tasks will match the tests where Fable 5 performs best. The pricing comparison is clearer than the business case: actual return depends on usage and workload, neither of which is captured by a model-wide score.

The source’s estimates for benchmark-run costs and blended token rates depend on stated usage assumptions. It also characterizes some comparisons with other models as directional because they rely on secondary reporting. Those figures should not be treated as a complete, independently verified account of every provider’s current pricing.

What Buyers Can Verify Next

The next meaningful evidence for buyers would be controlled, workload-specific evaluations that compare Fable 5 with lower-priced models on the same tasks. Useful tests would report completion quality, time spent by employees reviewing or correcting outputs, token use, and total cost—not only benchmark scores or a task’s elapsed time.

Companies considering the model can compare its results against Opus 4.8 using representative tasks and their own pricing mix, including any caching or batch use. Such a trial would not substitute for an independent study, but it could show whether the premium is justified for a particular deployment. Anthropic’s public launch account and any later customer evidence remain relevant, but the source material does not identify a forthcoming audited study or a scheduled next milestone.

Key Questions

How much does Claude Fable 5 cost?

Anthropic’s listed rates are $10 per million input tokens and $50 per million output tokens. The source report also lists discounted options, including prompt caching and batch pricing, subject to their applicable terms.

How much higher is Fable 5’s benchmark score than Opus 4.8’s?

Artificial Analysis reports an Intelligence Index score of 64.9 for Fable 5 and 61.4 for Opus 4.8, which the source report describes as a 5.7% difference. It is a benchmark comparison, not a direct measure of productivity in a company.

Is there a controlled study showing Fable 5 improves productivity?

The source report says it found no controlled human-baseline productivity study for Fable 5. The public example it identifies is Anthropic’s account of a Stripe migration, which it characterizes as an unaudited customer testimonial.

Does the Stripe example prove the migration took two months without Fable 5?

No independent proof is presented in the source material. Anthropic’s launch post says the work would otherwise have taken a team more than two months, but the report says the claim lacks a published methodology, independent audit, and replication.

Could Fable 5 still be worth the higher price?

Possibly, for workloads where its task-specific performance gains are valuable enough to offset the higher rates. The source report says Fable 5 led five of ten sub-benchmarks, but buyers would need to test representative tasks and calculate their own total costs.

Source: ThorstenMeyerAI.com