Twice the Price, 5.7% More Intelligence

Anthropic’s Claude Fable 5 is now listed at $10 per million input tokens and $50 per million output tokens, twice the price of Claude Opus 4.8, while benchmark data cited in a new report shows only a 5.7% gain on Artificial Analysis’s Intelligence Index. The gap matters for companies setting 2026 AI budgets, because the public case for large productivity gains remains limited.

The report says Fable 5’s pricing is confirmed by Anthropic’s launch materials, its platform pricing documents and product page, and is also reflected by outside coverage from Forbes, TechCrunch, Vellum, Finout and Artificial Analysis. The listed price makes Fable 5 the most expensive model in Anthropic’s active public Claude lineup.

By comparison, the report lists Claude Opus 4.8 at $5 per million input tokens and $25 per million output tokens. It lists Claude Sonnet 4.6 at about $3 per million input tokens and $15 per million output tokens, making Fable 5 more than three times as expensive as that model on sticker price.

The report also notes that the effective cost can be lower than the headline rate. With 90% prompt-caching discounts, cache reads at $1 per million tokens, batch pricing at $5 input and $25 output, and a 7:2:1 cache-hit-to-input-to-output blend, the cited blended rate is $7.70 per million tokens. US-only inference is listed at a 1.1x multiplier for data-residency needs.

Budget Pressure From Premium Pricing

The central issue is not whether Fable 5 performs better. The report says it does lead Opus 4.8 on aggregate benchmarks. The issue is whether a 2x price increase is justified by the public evidence now available to buyers.

Artificial Analysis data cited in the report puts Fable 5 at 64.9 on the Intelligence Index and Opus 4.8 at 61.4, a 5.7% increase. On GDPval-AA, described as a real-world knowledge-work Elo benchmark, Fable 5 is listed at 1,932 versus 1,890 for Opus 4.8, a 42-point lead.

That finding can cut in different directions. The report says the aggregate price-performance signal looks weak for many buyers, because a full Intelligence Index run is estimated at about $9,940 for Fable 5 versus $4,970 for Opus 4.8. But it also says Fable 5 set records in five of 10 sub-benchmarks, meaning the model may be more compelling for workloads where those specific gains appear.

Benchmarks Outpace Productivity Proof

The report separates two categories of evidence: pricing and benchmark data, which it describes as well documented, and enterprise productivity claims, which it says are much thinner. That distinction matters because model buyers often care less about benchmark scores than about measurable savings in staff time, output quality or delivery speed.

The main public productivity example cited is a Stripe story from Anthropic’s launch post. According to the report, Anthropic said Fable 5 handled a codebase-wide migration in a 50-million-line Ruby codebase in one day, work that Stripe said would otherwise have taken a whole team more than two months by hand.

The report treats that as a verified claim that was made, not as independently proven productivity data. It says no controlled human-baseline productivity studies were found in the public record, and it identifies only one unaudited enterprise proof point for human-baseline work gains.

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

— Anthropic launch post, as cited in the report

Productivity Gains Remain Unverified

It is still unclear whether Fable 5’s higher benchmark scores convert into broad workplace productivity gains large enough to offset its price. The report found no public controlled study comparing Fable 5-assisted work against a human or no-AI baseline across enterprise knowledge-work tasks.

The Stripe example may be meaningful, but the report says it remains a customer story, not an audited measurement. It does not establish how often similar gains occur, what setup was required, how much human review was involved, or whether the result applies outside that specific code migration.

Another open question is workload fit. The report says Fable 5’s advantage is stronger in some benchmark subcategories than in the blended score. Buyers still need to determine whether their own work resembles the sub-benchmarks where Fable 5 leads or areas where the gain is smaller.

Buyers Wait For Harder Evidence

The next test will be whether Anthropic, customers or independent evaluators publish controlled productivity studies that show Fable 5’s impact in real workflows. Until then, companies weighing the model face a decision based on confirmed price data, available benchmark scores and limited customer evidence.

For budget owners, the immediate step is likely workload-specific evaluation: comparing Fable 5, Opus 4.8 and cheaper models on internal tasks, with token costs and review time included. The report’s main finding is that the premium may be defensible for some hard tasks, but the public evidence does not yet prove a broad productivity return.

Key Questions

How much does Claude Fable 5 cost?

The report lists Claude Fable 5 at $10 per million input tokens and $50 per million output tokens. It says batch pricing and prompt caching can reduce effective cost for some usage patterns.

How does Fable 5 compare with Claude Opus 4.8?

According to the cited benchmark data, Fable 5 scored 64.9 on the Artificial Analysis Intelligence Index, compared with 61.4 for Opus 4.8. Opus 4.8 is listed at half the price of Fable 5.

Is Fable 5 proven to make workers more productive?

The report says no public controlled human-baseline studies were found. The main public productivity example is an unaudited Stripe customer story cited by Anthropic.

Could Fable 5 still be worth the premium?

Possibly. The report says Fable 5 leads some sub-benchmarks by more than the aggregate score suggests. The business case depends on whether a buyer’s own workload matches those areas and whether the gains exceed the higher token cost.

Source: Thorsten Meyer AI