🔍 Read the full analysis: Claude Fable 5.2 Task Costs: A Breakdown Of Its Effort Options on ThorstenMeyerAI.com
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
TL;DR
Anthropic released Claude Fable 5.2 on November 9, 2026, with five effort settings that trade task cost and speed for performance on an independent AI benchmark. Artificial Analysis reports costs of $3.10 per task at Low effort and $26.54 at Max, while the score rises from 54 to 65; the benchmark does not show that Max is best for every task.
Anthropic released Claude Fable 5.2 on November 9, and Artificial Analysis’s benchmark puts its five effort settings at $3.10 to $26.54 per task, with scores from 54 to 65. The figures give businesses a concrete way to compare the cost, speed and benchmark performance of asking the model to spend less or more effort before returning an answer.
Artificial Analysis tested the model on its Intelligence Index, which covers nine evaluations spanning agentic coding, business documents, science and knowledge. It reports a score of 54 for Low effort, 56 for Medium, 62 for High, 62 for Xhigh and 65 for Max. Among the 74 models in the index, those settings ranked 15th, ninth, third, second and first, respectively. The scores are rounded as displayed by Artificial Analysis.
The reported average cost per index task rises from $3.10 at Low to $3.80 at Medium, $9.08 at High, $14.83 at Xhigh and $26.54 at Max. Average task time also increases, from 6.5 minutes at Low to 46.4 minutes at Max. Artificial Analysis’s cost figure is an average API cost for one task in its test, not a guaranteed charge for every user request.
Anthropic lists the same standard token prices across effort settings: $8 per million input tokens and $40 per million output tokens. The main difference in benchmark spending is the amount of output used while the model reasons: Artificial Analysis reports 34.9 million output tokens across its full index run at Low and 275 million at Max. Anthropic also offers an optional faster mode, priced at $16 per million input tokens and $80 per million output tokens, which it says can generate tokens up to 2.5 times faster.
Task costs rise with effort but results depend on the job
Artificial Analysis measured five effort settings on its Intelligence Index. Higher effort increased benchmark scores overall, but also raised average cost and task time—and Max was not best on every evaluation.
Cost and score by effort
Benchmark averages show the price of asking the model to spend more effort.
| Setting | Cost per task | Score | Index rank |
|---|---|---|---|
| Low | $3.10 | 54 | 15th |
| Medium | $3.80 | 56 | 9th |
| High | $9.08 | 62 | 3rd |
| Xhigh | $14.83 | 62 | 2nd |
| Max | $26.54 | 65 | 1st |
What higher effort changes
More effort generally means more output tokens and a longer wait in this test.
- Lower benchmark task costs
- Low averaged 6.5 minutes per task
- May suit work where speed or cost matters most
- Higher scores overall on the index
- Max averaged 46.4 minutes per task
- May suit demanding work where performance is a priority
A higher setting is not always better
The index is a reference point, not a guarantee for a company’s own workload.
Max did not lead every test
On one cited coding evaluation, High scored 49.1%, slightly above Max at 48.8%.
Test your own tasks
Compare quality, response time and token cost on representative work before choosing a setting.
Not a fixed request price
Reported costs are averages from the index test; actual charges depend on input and output usage.
A practical starting point
Anthropic’s recommendation is to begin with Claude Opus 5.5 for most coding and knowledge work, then use Fable when a task calls for it.
Include the kinds of coding, research or document tasks your team actually handles.
Record answer quality, response time and token cost for each setting.
Use higher effort where your results justify the additional time and cost.
The Cost of More Model Effort
The benchmark illustrates that choosing an effort setting is a cost-and-time decision, not simply a choice between a capable and incapable model. Moving from Medium to High raises the reported score by six points and the average task cost by $5.28. Moving from High to Xhigh costs a further $5.75 per task while the displayed score remains 62. Max adds three points over High, but costs nearly three times as much and takes considerably longer on average.
Those comparisons may help organizations deciding how to allocate model use across routine and demanding work. A lower setting could be sufficient for tasks where speed or cost matters most, while higher effort may be worth testing for work where benchmark performance is a priority. But the results do not establish which setting performs best on a particular company’s own jobs. In one Artificial Analysis coding evaluation cited in the source, Fable 5.2 at High scored 49.1%, slightly above its 48.8% at Max, a reminder that more effort does not guarantee a better result on every evaluation.
Artificial Analysis also reports that Fable 5.2 at High scored 62 for $9.08 per task, compared with Claude Opus 5.5 at Max, which scored 59 for $10.04. That is a useful comparison on this index, not evidence that Fable will outperform Opus on all tasks, or that the measured prices will match a reader’s usage.
Popular right now
Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Bluetooth 5.3, USB-C Charging, and IPX4 Water Resistance
Vitamix 5200 Professional-Grade Blender
Dyson V8 Motorhead Origin Cordless Vacuum
As an affiliate, we earn on qualifying purchases.
Fable and Mythos Release Details
Anthropic says Fable 5.2 and Claude Mythos 5.2 replace their respective 5.1 versions. The company describes Mythos as the same underlying model as Fable, but without some of Fable’s safety restrictions; access to Mythos is limited to organizations accepted into Anthropic’s Cyber or Life Sciences Verification Programs. Fable is available to users with safeguards intended to block certain high-risk cybersecurity work and most research-biology tasks.
According to Anthropic, when Fable blocks a query, it returns an answer from an older model: Opus 4.8 for cybersecurity and Opus 5 for biology. The company says early testing found about 55% fewer blocks than with Fable 5.1. That is an Anthropic-reported result; the source does not provide independent verification of the testing or further detail about its methodology.
Anthropic recommends starting with Claude Opus 5.5 for most coding and knowledge work, then moving to Fable when a task calls for it. The company also says the 5.2 models write answers up to 30% faster than the 5.1 versions. Its release materials describe demanding prerelease demonstrations, including a multi-agent attempt to formalize Fermat’s Last Theorem. Those examples are company accounts and should not be treated as independently verified measures of ordinary user performance.
“For most coding and knowledge work, start with Claude Opus 5.5 and move up to Fable only when the task needs it.”
— Anthropic
Limits of the Benchmark Evidence
The reported costs and rankings describe Artificial Analysis’s test setup, not every workload or a customer’s expected monthly bill. The supplied information does not give enough detail to determine how closely the index tasks match a particular business’s coding, research or document work. The rounded scores also leave open whether small differences between settings are meaningful beyond this test.
Anthropic’s claims about speed, safety blocks and prerelease demonstrations are not independently verified in the source material. Nor does the benchmark establish that the highest-scoring setting is the most useful or economical choice for a given task. Actual costs will depend on usage, including how much input and output the model consumes.
The source article also disclosed that it was drafted with help from Claude, that Anthropic is the subject, and that neither Anthropic nor Artificial Analysis reviewed or sponsored it. That disclosure is relevant to readers weighing the article’s framing, alongside the distinction between independently reported benchmark figures and company claims.
Test Settings on Real Workloads
The immediate next step for prospective users is to compare effort levels on their own representative tasks, recording quality, response time and token cost rather than relying on the index score alone. Anthropic’s recommendation is to begin with Opus 5.5 for most coding and knowledge work and use Fable where the task warrants it. The source material does not announce a later benchmark release or a further product milestone.
As more users and evaluators publish results, comparisons across real-world workloads may clarify whether the cost increases at High, Xhigh and Max translate into better outcomes for particular jobs. Until then, the figures provide a benchmark-based price and performance reference, not a universal setting recommendation.
Key Questions
What does an effort level change?
Fable 5.2 offers Low, Medium, High, Xhigh and Max. In Artificial Analysis’s test, higher settings generally used more output tokens and took longer, with benchmark scores varying by setting.
How much does a Fable 5.2 task cost?
Artificial Analysis reports average costs of $3.10 at Low, $3.80 at Medium, $9.08 at High, $14.83 at Xhigh and $26.54 at Max for one task in its Intelligence Index. These are benchmark averages, not fixed prices for every request.
Is Max the best setting for every task?
No. Max had the highest score on the Intelligence Index, but the supplied source says Fable 5.2 at High slightly outscored Max on an Artificial Analysis coding evaluation. Results can vary by task, and Max was also the slowest and most expensive setting in the index figures.
How is Mythos 5.2 different from Fable 5.2?
Anthropic describes Mythos as the same underlying model without some of Fable’s safety restrictions. The company says Mythos is limited to organizations accepted into its Cyber or Life Sciences Verification Programs; Fable is available more broadly with safeguards.
Are the reported task costs guaranteed API charges?
No. They are Artificial Analysis’s average API costs for tasks in its benchmark. Anthropic’s stated standard token prices are $8 per million input tokens and $40 per million output tokens, while the final bill depends on actual token use and any selected fast mode.
Source: ThorstenMeyerAI.com
