🔍 Read the full analysis: What Does Each Claude Fable 5.2 Effort Level Cost Per Task? 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
Artificial Analysis reports that Claude Fable 5.2 costs an average of $3.10 per Intelligence Index task at Low effort and $26.54 at Max. Max scored 65, compared with 54 at Low, but took more than seven times as long on average; the figures are benchmark results, not guaranteed costs for every task.
Claude Fable 5.2’s estimated cost per task ranges from $3.10 at Low effort to $26.54 at Max in Artificial Analysis’s Intelligence Index test, according to results accessed November 10. The benchmark puts Max at 65 points and No. 1 among 74 models, while Low scores 54, giving businesses a measured trade-off between benchmark performance, API spending and time.
Artificial Analysis evaluated the model at five settings: Low, Medium, High, Xhigh and Max. Its reported cost per task is the average API cost for one task in the Intelligence Index evaluation, which covers nine tests across agentic coding, business documents, science and knowledge. The figures should not be read as a fixed price for any particular customer request.
The reported results rise from 54 at Low for $3.10 and 56 at Medium for $3.80 to 62 at High for $9.08. Xhigh also scores 62, at $14.83, while Max reaches 65 for $26.54. Average task time increases alongside effort: from 6.5 minutes at Low to 46.4 minutes at Max. Scores are rounded as displayed by Artificial Analysis.
Anthropic lists the same standard token rates for every effort setting: $8 per million input tokens and $40 per million output tokens. The benchmark’s cost differences reflect how much the model generates while working, not a different listed rate at each setting. Artificial Analysis counted 34.9 million output tokens across the index at Low and 275 million at Max.
What Fable 5.2 costs at each effort level
Artificial Analysis estimates a wide cost and runtime trade-off across five effort settings. These are benchmark averages, not guaranteed prices for individual tasks.
Cost rises with effort
Each estimate is an average API cost for one Intelligence Index task.
What extra effort buys
Higher settings do not guarantee better results on every task.
A measured step up
High scores 62 for $9.08 per task, up from Medium’s 56 for $3.80.
More time, modest score gain
Max adds three points over High, costs nearly three times as much, and takes 46.4 minutes on average.
High edged Max
On FrontierCode 2.0, High scored 49.1% and Max scored 48.8%.
Same listed token rates
The benchmark cost gap reflects usage, not a different standard rate for each setting.
- $8 per million input tokens at every effort level
- $40 per million output tokens at every effort level
- 34.9 million output tokens at Low
- 275 million output tokens at Max
Make the comparison relevant
The index covers nine tests across coding, business documents, science and knowledge.
Compare settings on work your team actually does.
Measure quality, token use, latency and total cost.
Benchmark estimates do not determine your production bill.
The Cost of More Model Effort
The results give companies a concrete, though benchmark-specific, way to weigh extra reasoning against cost and waiting time. Moving from Medium to High raises the index score from 56 to 62, while the estimated task cost increases from $3.80 to $9.08 and average time rises from 8.3 to 17.9 minutes. That could matter for teams running many requests or workflows where people are waiting for a result.
The upper settings do not show the same score gain per additional dollar in this test. Xhigh costs $14.83 and records the same rounded score as High. Max adds three points over High but costs nearly three times as much and takes 46.4 minutes on average. These figures do not prove that High is best for every use: performance can vary by task, and scores rounded to whole numbers may conceal smaller differences.
The benchmark also cautions against treating the highest setting as automatically superior. On Artificial Analysis’s FrontierCode 2.0 agentic coding test, Fable 5.2 at High scored 49.1%, slightly above its 48.8% result at Max. The result is specific to that test, but it shows why a business should compare settings on its own workload rather than assume more effort always improves outcomes.
Products Worth Considering
Apple MacBook Pro 16-inch (2021), M1 Pro, 10-core CPU, 16-core GPU, 16GB RAM, 512GB SSD, Silver
Logitech MX Master 4 Ergonomic Wireless Mouse with Haptic Feedback, Graphite
Ledger Nano X Bluetooth Crypto Wallet
As an affiliate, we earn on qualifying purchases.
How the Benchmark Fits Anthropic’s Launch
Anthropic announced Claude Fable 5.2 and Claude Mythos 5.2 on November 9, describing them as replacements for their 5.1 versions. The source article says Fable is available to users with strong safeguards for high-risk cybersecurity and much research biology work. Mythos is described as the same underlying model without those blocks, available only to organizations accepted into Anthropic’s Cyber or Life Sciences Verification Programs. These access and safety details are Anthropic’s descriptions.
Anthropic says the new models produce answers up to 30% faster than their 5.1 versions and offers an optional fast mode with token rates of $16 per million input tokens and $80 per million output tokens. The effort-level cost figures above come instead from Artificial Analysis’s benchmark and use the standard listed rates. Anthropic’s separate claims about pre-release demonstrations, including a multi-agent attempt to formalize Fermat’s Last Theorem, are vendor-reported results and have not been independently verified in the supplied material.
For comparison, the source reports that Opus 5.5 at Max scored 59 at an estimated $10.04 per task on the same index. Fable 5.2 at High scored 62 at $9.08. Those results provide a benchmark comparison, not a promise that Fable will be cheaper or perform better on every business’s tasks.
“For most coding and knowledge work, Anthropic recommends starting with Claude Opus 5.5 and moving up to Fable only when the task needs it.”
— Anthropic
Limits of the Cost Estimates
The $3.10 to $26.54 figures are averages from one benchmark suite, not prices guaranteed for every prompt or production workflow. Actual API bills depend on token use and task characteristics; the supplied results do not establish what a particular company will spend across its own workload. The material also does not provide uncertainty ranges for the benchmark scores or costs.
It remains unclear how well the ranking and effort-level differences generalize to other evaluations, languages, task mixes or real-world deployments. The scores are rounded, and the supplied account does not include enough detail to assess every testing choice independently. Anthropic’s pre-release performance and safety statements are company claims unless separately verified; the source notes that the Fermat results, for example, were not independently verified.
Test Settings on Real Workloads
The next practical step for organizations considering Fable is to compare effort levels on representative tasks, tracking quality, token use, latency and total cost rather than relying on the index alone. Anthropic’s reported recommendation is to begin with Opus 5.5 for most coding and knowledge work, then use Fable where the task calls for it. No further release milestone or independent follow-up test is specified in the supplied material.
Organizations considering Mythos also face a separate access question: Anthropic says use is limited to approved participants in its verification programs. The source does not state how many organizations have been accepted or provide a timeline for broader access. API charges and model behavior can change, so published benchmark estimates should be checked against current pricing and actual usage.
Key Questions
How much does Claude Fable 5.2 cost per task?
In Artificial Analysis’s Intelligence Index test, the estimated average ranges from $3.10 at Low effort to $26.54 at Max. These are benchmark task averages, not fixed prices for every request.
Which effort level scored highest?
Max scored 65 in the reported Intelligence Index results, placing first among 74 models. It also had the highest estimated task cost and longest average runtime among the five settings.
Is High effort a better value than Max?
In this benchmark, High scored 62 for $9.08 per task, while Max scored 65 for $26.54. That makes High less expensive in this test, but results and costs may differ on a company’s own tasks.
Do all effort settings have different token prices?
No. The source reports standard rates of $8 per million input tokens and $40 per million output tokens across the settings. The benchmark estimates different task costs because the model uses different amounts of output tokens as effort changes.
Are the benchmark task costs guaranteed API bills?
No. They are Artificial Analysis estimates for its test tasks. A real request’s cost depends on its token use and other details; the supplied material does not give a guaranteed cost for individual customer workloads.
Source: ThorstenMeyerAI.com
