TL;DR
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Anthropic released Claude Fable 5.2 on November 9, and Artificial Analysis ranked it first among 74 models at Max effort the following day. The index score rises from 54 at Low to 65 at Max, while average task cost rises from $3.10 to $26.54; High scores 62 for $9.08.
Anthropic released Claude Fable 5.2 on November 9, and independent model evaluator Artificial Analysis ranked it first on its Intelligence Index at Max effort a day later. The result comes with a large cost range: the benchmark estimates $3.10 per task at Low and $26.54 at Max, while scores rise from 54 to 65.
Artificial Analysis evaluated five effort settings across nine assessments covering agentic coding, business documents, science and knowledge. It reports an average API cost per task for its index, rather than a universal price for every use. Its figures show Low at 54 points and $3.10 per task; Medium, the default in Claude apps and Claude Code, at 56 and $3.80; and High at 62 and $9.08.
The higher settings cost more without adding points at every step. Xhigh scores 62, the same rounded score as High, at $14.83 per task. Max scores 65 at $26.54 per task and takes an average 46.4 minutes per task, compared with 17.9 minutes at High and 6.5 minutes at Low. Artificial Analysis reports that total output tokens across its index run grow from 34.9 million at Low to 275 million at Max.
Anthropic lists the same API rates for all five effort levels: $8 per million input tokens and $40 per million output tokens. The benchmark’s different estimated bills reflect differing token use, including the model’s additional reasoning output. Anthropic also offers an optional fast mode at higher rates—$16 per million input tokens and $80 per million output tokens—and says Fable 5.2 generates answers up to 30% faster than version 5.1.
Is Claude Fable 5.2 Worth It? Cost per task by effort level
Artificial Analysis ranked Fable 5.2 first at Max effort, but the highest score comes with a much higher estimated task cost. The best setting depends on the work and the value of extra performance.
Cost rises with effort
Artificial Analysis estimates average cost across its index tasks; actual bills vary with token use.
| Effort | Index score | Cost per task |
|---|---|---|
| Low | 54 | $3.10 |
| Medium | 56 | $3.80 |
| High | 62 | $9.08 |
| Xhigh | 62 | $14.83 |
| Max | 65 | $26.54 |
More effort does not always score higher
The index rewards Max overall, but results vary by assessment.
- 62 index points for $9.08 per task.
- Scored 49.1% on FrontierCode 2.0 coding.
- Average task time: 17.9 minutes.
- 65 index points for $26.54 per task.
- Scored 48.8% on FrontierCode 2.0 coding.
- Average task time: 46.4 minutes.
What changes the bill?
Standard API token rates are the same across all five effort levels.
$8 per million input tokens and $40 per million output tokens.
Index output grows from 34.9 million tokens at Low to 275 million at Max.
Average time rises from 6.5 minutes at Low to 46.4 minutes at Max.
Choose for the workload
Medium is the default in Claude apps and Claude Code; benchmark results are a starting point, not a guarantee for your tasks.
- Compare settings on representative work.
- Track quality, response time and API cost together.
- Use higher effort when the added performance matters.
The Cost of Higher Effort
For organizations choosing a model setting, the figures show why the top benchmark rank does not by itself identify the best value for routine work. High scores 62 for $9.08, while Max costs nearly three times as much for three additional index points. That difference may matter for teams running many tasks, especially when they also need to account for longer response times.
Results can depend on the work being measured. On Artificial Analysis’s FrontierCode 2.0 agentic coding test, Fable 5.2 at High scored 49.1%, slightly above its Max result of 48.8%. The index and this coding test measure different things, so neither figure establishes how the model will perform on a particular company’s tasks. They do show that more effort did not improve every reported score.
A comparison with Anthropic’s Claude Opus 5.5 offers another reference point. Artificial Analysis lists Opus 5.5 at Max effort at 59 points and $10.04 per index task. Fable 5.2 at High scores three points higher for about 10% less per task in that evaluation, even though its listed token rates are twice Opus 5.5’s. These are benchmark estimates, not guaranteed savings; actual costs depend on prompts, outputs and usage.
Products Worth Considering
Bose QuietComfort Wireless Noise-Cancelling Over-Ear Headphones, Twilight Blue
Apple AirPods Pro (2nd Generation) Wireless Earbuds with USB‑C Charging, Up to 2X More Active Noise Cancelling, Transparency Mode, Adaptive
Sony WH-1000XM6 Wireless Noise-Canceling Headphones, Black
As an affiliate, we earn on qualifying purchases.
Fable and Mythos Access
Anthropic released Fable 5.2 and Claude Mythos 5.2 as replacements for their respective 5.1 versions. Anthropic describes Mythos as the same underlying model as Fable, made available to organizations accepted into its Cyber or Life Sciences Verification Programs. Fable is available to users generally and has safeguards aimed at high-risk cybersecurity work and many research biology tasks.
According to Anthropic, when Fable blocks a query in those areas, an older model may respond instead: 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 test result; the supplied material does not provide an independent evaluation of the change.
Anthropic has also described prerelease demonstrations involving electronic-device designs, black-hole physics problems and a proof formalization project involving many agents. Those are company accounts of selected tests, not results from the Artificial Analysis index or evidence of typical performance in everyday work. Artificial Analysis’s published scores and task-cost estimates provide a separate, standardized comparison, with their own limits.
““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 Cost Estimates
The $3.10 to $26.54 figures are benchmark averages for Artificial Analysis’s index tasks, not fixed charges for any request. The supplied material does not provide a task-by-task breakdown, and a business’s costs will vary with its prompts, output length, chosen effort level and use of fast mode. The index score also does not predict performance on every coding, research or office task.
The reported scores are rounded, so High and Xhigh both appear as 62 despite their different costs. The material does not establish whether the difference between those settings would be meaningful for a particular workload. Anthropic’s claims about faster answers, fewer safety blocks, demonstrations and coding costs are company statements; the details provided do not include independent confirmation of those claims.
Choosing a Setting for Work
Teams considering Fable 5.2 can compare effort levels on representative tasks and track quality, latency and API cost together. The published benchmark points to High as a lower-cost setting with a score close to the top results, while Max is the most expensive and slowest listed option. Whether that added effort is worthwhile depends on the task and the value of the additional performance.
Further evidence could come from independent evaluations, more detailed task-level results and customer usage data. Until those are available, the Artificial Analysis results offer a dated comparison across its test set, while Anthropic’s recommendations and product claims remain attributable to the company.
Key Questions
Which Fable 5.2 effort level costs the least per benchmark task?
Low is the least expensive setting in Artificial Analysis’s index, at an estimated $3.10 per task. Its score is 54.
Is Max effort worth the higher cost?
The benchmark does not settle that for every use. Max scores 65 for $26.54 per index task, while High scores 62 for $9.08. The value depends on whether the extra performance matters for a given workload.
What is the default Fable 5.2 effort setting?
Medium is the default in Claude apps and Claude Code, according to the source material. Artificial Analysis reports a score of 56 and an average index task cost of $3.80 at Medium.
Do all effort levels have different token prices?
No. Anthropic lists standard rates of $8 per million input tokens and $40 per million output tokens across the effort levels. The estimated task costs differ because token usage varies by setting.
Source: Thorsten Meyer AI
