🔍 Read the full analysis: How To Choose A Claude Fable 5.2 Effort Level Based On Task Cost on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Anthropic released Claude Fable 5.2 on November 9, and Artificial Analysis testing shows that its five effort levels trade higher scores for higher cost and longer task times. High scored 62 for $9.08 per test task, while Max scored 65 for $26.54; the results suggest the top setting may not be worth the added cost for every job.
Anthropic’s Claude Fable 5.2, released November 9, has five selectable effort levels that produce sharply different benchmark costs: Artificial Analysis measured $3.10 per task at Low and $26.54 at Max. The higher-cost setting scored 65 on the index, compared with 54 at Low, giving businesses a concrete trade-off between performance, spending and time before choosing how much reasoning to buy.
Artificial Analysis tested Fable 5.2 across nine evaluations covering agentic coding, business documents, science and knowledge. Its published figures show scores and average costs per index task of 54 and $3.10 at Low; 56 and $3.80 at Medium; 62 and $9.08 at High; 62 and $14.83 at Xhigh; and 65 and $26.54 at Max. The ranking is among 74 models: Max placed first, Xhigh second and High third, though the displayed scores are rounded.
The benchmark points to a substantial cost increase between Medium and High, which adds six score points for $5.28 more per task. Xhigh costs about 63% more than High while showing the same rounded score. Max adds three points over High but costs almost three times as much. Average task time also rises, from 6.5 minutes at Low to 46.4 minutes at Max. These are test results, not a guarantee that a particular business task will improve by the same amount.
Anthropic lists Fable 5.2 at $8 per million input tokens and $40 per million output tokens, regardless of effort level. The bill changes because higher effort can use more tokens: Artificial Analysis reported 34.9 million output tokens across the full index at Low and 275 million at Max. Medium is the default in Claude apps and Claude Code. Anthropic also offers an optional fast mode at higher token prices, but the supplied benchmark cost figures describe the tested effort levels rather than establishing a universal cost for every user’s workload.
Choose effort by task cost and measured need
Claude Fable 5.2’s effort levels trade benchmark score for cost and time. The best setting depends on what your work actually requires.
What the benchmark found
Artificial Analysis tested five settings across nine evaluations. These are averages from its index, not guaranteed results for your workload.
| Effort | Score | Cost per task |
|---|---|---|
| Low | 54 | $3.10 |
| Medium | 56 | $3.80 |
| High | 62 | $9.08 |
| Xhigh | 62 | $14.83 |
| Max | 65 | $26.54 |
How to choose a starting point
Match effort to the task, then check whether added quality is worth the extra spend and wait.
Medium is the default in Claude apps and Claude Code; Low was the cheapest tested setting.
Compare task quality, token use, cost and elapsed time on work your team actually handles.
Use a higher setting only where observed performance gains justify the added cost and time.
Where the trade-off sharpens
Higher settings do not guarantee better results on every task, and equal rounded scores do not prove identical performance.
- Score 62 at $9.08 per test task
- FrontierCode 2.0 score: 49.1%
- Score 65 at $26.54 per test task
- FrontierCode 2.0 score: 48.8%
The Cost of More Reasoning
The results matter to companies setting budgets for repeated AI work. A small per-task difference can accumulate across high-volume use, while a longer wait at Max may affect workflows that depend on quick responses. The index provides a common comparison, but actual costs will depend on the task, prompts, output and how often the model is used.
For teams choosing a starting point, High appears to be the benchmark’s middle ground: it scored 62 at $9.08 per task, compared with 65 at $26.54 for Max. That does not establish High as best for every use case. Artificial Analysis also reported that Fable 5.2 at High scored 49.1% on its FrontierCode 2.0 coding test, slightly above Max at 48.8%. Results vary by evaluation, so a higher effort setting should be tested against the work it is meant to handle.
The source article reports that Anthropic recommends starting with Claude Opus 5.5 for most coding and knowledge work, then moving to Fable when a task requires it. In Artificial Analysis’s index, Opus 5.5 at Max scored 59 at $10.04 per task, while Fable at High scored 62 at $9.08. That comparison favors Fable on those reported figures, but does not establish the best choice across all workloads.
Popular right now
De'Longhi Rivelia Automatic Espresso Machine with Grinder and Milk Frother, Beige
Ninja Crispi Microwave & Air Fryer with FusionCrisp Technology, 14-in-1, Black (MW101BK1)
KitchenAid Artisan 5-Qt Tilt Head Stand Mixer, Milkshake
As an affiliate, we earn on qualifying purchases.
Fable’s Five Effort Settings
Fable 5.2 replaced Fable 5.1 on November 9, according to the supplied source. Anthropic says the new model generates answers up to 30% faster than its predecessor. Fable has five settings—Low, Medium, High, Xhigh and Max—that let users adjust how much effort the model applies. The reported index costs and times are averages from Artificial Analysis’s test, not fixed prices for every request.
The release also includes Claude Mythos 5.2, described in the source as the same underlying model with different safety restrictions. Fable is available to users with safeguards that block some high-risk cybersecurity and biology requests; Mythos is limited to organizations accepted into Anthropic’s Cyber or Life Sciences Verification Programs. These access and safety details are claims attributed to Anthropic in the source material, not independent benchmark findings.
The source also describes vendor-reported demonstrations, including work on electronic devices and mathematical proofs. Anthropic’s accounts of those demonstrations are not independent verification of typical performance. They provide release context but do not answer the practical question measured by Artificial Analysis: how benchmark score, time and average task cost change with effort.
Limits of the Benchmark Results
The figures do not establish what effort level is most cost-effective for an individual company. Artificial Analysis’s cost per task is an average from its own index; the supplied material does not provide a matching estimate for each reader’s prompts, usage volume or production workflow. Scores are rounded, and equal displayed scores do not prove that two settings perform identically.
The source material does not specify whether the reported benchmark costs include every possible operational expense, or how results translate to particular live tasks. Anthropic’s safety statements, speed claims and accounts of prerelease demonstrations are company-reported. The demonstrations described have not been independently verified in the supplied material. It is also unclear how performance and costs will change as more independent evaluations become available.
Test Settings Against Real Work
Organizations considering Fable can compare effort levels on a small set of representative tasks before deploying them broadly. They can record task quality, token use, cost and elapsed time, then set different effort levels for routine requests and work where an error would be more consequential. That approach would show whether the benchmark’s gains apply to their own workloads.
Further independent testing may clarify how Fable 5.2 performs across different task categories and whether its cost rankings hold outside the Artificial Analysis index. Until then, the available figures support a measured choice: begin with a lower or middle setting, compare results against the work at hand, and pay for higher effort only where observed performance justifies it.
Key Questions
Which Fable 5.2 effort level is cheapest?
Low was cheapest in Artificial Analysis’s index, averaging $3.10 per task. It scored 54 in that test; the figure is not a guaranteed cost for every request.
Does Max perform much better than High?
On the index, Max scored 65 and High scored 62, with displayed scores rounded. Max cost $26.54 per task versus $9.08 at High and took longer on average.
Why does effort level change the cost if token prices are the same?
The listed per-token prices are the same across effort settings, but higher effort can use more output tokens. Artificial Analysis reported 34.9 million output tokens across the index at Low and 275 million at Max.
Is High the best setting for coding?
The supplied source reports that High scored 49.1% on Artificial Analysis’s FrontierCode 2.0 test, slightly above Max at 48.8%. That result is limited to the named evaluation and does not establish which setting is best for other coding tasks.
How should a business choose an effort level?
Compare settings on representative work, tracking quality, time and cost. Anthropic’s reported guidance is to start with Opus 5.5 for most coding and knowledge work and use Fable when needed; the best choice depends on the task and measured results.
Source: ThorstenMeyerAI.com
