🔍 Read the full analysis: Claude Fable 5.2’S New #1 Status Comes With Different Costs Per Task on ThorstenMeyerAI.com
Get smart everyday buys 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 Nov. 9, and Artificial Analysis ranked it first among 74 models at Max effort, scoring 65. On that benchmark, a task costs $26.54 at Max versus $9.08 at High, which scored 62; the figures reflect the test’s average API costs, not a fixed charge for every real-world task.
Anthropic’s Claude Fable 5.2 ranked No. 1 among 74 models on the Artificial Analysis Intelligence Index at its Max effort setting, but the benchmark’s reported average cost was $26.54 per task. At High effort, Fable scored 62 rather than 65 and cost $9.08 per task, highlighting a trade-off businesses face when choosing how much computation to use.
Artificial Analysis tested Fable across five effort levels on an index covering nine evaluations in areas including agentic coding, business documents, science and knowledge. Its reported score rose from 54 at Low to 65 at Max. Medium, the default in Claude apps and Claude Code, scored 56; High and Xhigh each displayed a rounded score of 62.
The index listed average task costs of $3.10 at Low, $3.80 at Medium, $9.08 at High, $14.83 at Xhigh and $26.54 at Max. Average task time also climbed, from 6.5 minutes at Low to 46.4 minutes at Max. These are Artificial Analysis figures for its test, not a universal quote for every user’s workload.
Anthropic lists Fable 5.2 API pricing at $8 per million input tokens and $40 per million output tokens across effort settings. The source attributes the higher benchmark costs at greater effort to the model generating more output tokens while reasoning: 34.9 million across the index at Low and 275 million at Max. Anthropic also offers an optional fast mode at higher token prices, $16 per million input tokens and $80 per million output tokens.
Claude Fable 5.2 ranks first at a higher task cost
Artificial Analysis ranked Fable first at Max effort, but its benchmark shows a steep trade-off between score, API cost and completion time. These are test averages, not fixed prices for real-world tasks.
Effort changes the trade-off
Higher effort raised the index score, but benchmark costs and time also increased.
| Effort | Score | Average task cost | Average time |
|---|---|---|---|
| Low | 54 | $3.10 | 6.5 minutes |
| Medium | 56 | $3.80 | Not stated |
| High | 62 | $9.08 | 17.9 minutes |
| Xhigh | 62 | $14.83 | Not stated |
| Max | 65 | $26.54 | 46.4 minutes |
A top score is not a universal winner
Results vary by evaluation, so the ranking alone cannot predict performance on a company’s own work.
- Intelligence Index score: 65
- Average index task cost: $26.54
- FrontierCode 2.0: 48.8%
- Intelligence Index score: 62
- Average index task cost: $9.08
- FrontierCode 2.0: 49.1%
Two versions, different access
The supplied account describes Fable and Mythos as sharing an underlying model, with different availability and safeguards.
Fable 5.2
Available to users generally, with safeguards for certain high-risk cybersecurity and biology requests.
Mythos 5.2
Available only to organizations accepted into Anthropic’s Cyber or Life Sciences Verification Programs.
Test before choosing a setting
Compare quality, spending and completion time on the tasks your team actually runs.
Use representative prompts and workflows from your own production work.
Measure output quality alongside API cost and time to completion.
Use higher effort when its value justifies the added cost and delay.
The Cost of Fable’s Top Score
The results make the headline ranking less useful on its own for buyers comparing models. A higher effort setting may improve results on some tasks, but the added score comes with higher cost and longer waits. In this index, moving from High to Max added three displayed points, while the reported average task cost rose from $9.08 to $26.54 and average time increased from 17.9 to 46.4 minutes.
The relationship is not a simple guarantee that more effort produces better results. Artificial Analysis’s FrontierCode 2.0 coding test gave Fable 48.8% at Max and 49.1% at High, according to the supplied material. That is one evaluation, not proof that High is generally better, but it illustrates why teams may need to test settings against their own tasks rather than choose Max by default.
The same source compares Fable at High with Anthropic’s Claude Opus 5.5 at Max: Fable scored 62 for $9.08 per index task, while Opus scored 59 for $10.04. That comparison is specific to Artificial Analysis’s benchmark and its cost method. It does not establish that Fable will be cheaper or perform better across all business workloads.
Products Worth Considering
Samsung 32-inch Odyssey OLED G8 (G80SH) 4K Gaming Monitor
Dell UltraSharp U4025QW 40 5K2K WUHD Curved LED Monitor (21:9)
Dell UltraSharp U2424H 24-Inch FHD Monitor
As an affiliate, we earn on qualifying purchases.
Two Names, Different Access Rules
Anthropic released Fable 5.2 and Mythos 5.2 on Nov. 9, replacing their 5.1 versions. According to the supplied account, they share the same underlying model, while Mythos is available only to organizations accepted into Anthropic’s Cyber or Life Sciences Verification Programs. Fable is available to users generally and retains safeguards intended to block certain high-risk cybersecurity work and most research biology tasks.
Anthropic says a blocked Fable query receives a response from an older model—Opus 4.8 for cybersecurity and Opus 5 for biology—and that early testing found about 55% fewer blocks than with Fable 5.1. The company also says the new models generate answers up to 30% faster than their predecessors. These are vendor statements, not findings established by the Artificial Analysis ranking.
Anthropic’s recommendation, as described in the source, is to start most coding and knowledge-work tasks with Claude Opus 5.5 and move to Fable when a task requires it. The source also recounts ambitious prerelease demonstrations, including a multi-agent attempt to formalize Fermat’s Last Theorem. Anthropic reported that Fable proved all weighted statements in that run after 30 hours, while Opus 5.5 reached 27% by that point. The account says each model had one run with about 1,000 agents; these are Anthropic’s results and have not been independently verified.
“For most coding and knowledge work, start with Claude Opus 5.5 and move up to Fable only when the task needs it.”
— Anthropic, as quoted in the supplied source
Limits of the Benchmark Evidence
The ranking and cost estimates describe Artificial Analysis’s Intelligence Index, which uses nine evaluations; they do not predict the bill or performance for every company’s prompts. The supplied material does not specify how representative those tasks are of particular production workloads, or how results may vary with prompt length, tool use and other operating choices.
Displayed scores are rounded, so High and Xhigh both appear as 62 even though the source ranks them third and second, respectively. The material does not explain the difference behind those ranks. It also does not provide independent verification of Anthropic’s safety-block comparison, speed claim or demonstration results. The reported costs should be read as averages for the index test, not fixed per-task prices.
Testing Settings on Real Workloads
The immediate next step for teams evaluating Fable is to compare effort levels on their own tasks, tracking quality, API spending and completion time together. The benchmark suggests High may offer a different balance from Max, but the best setting will depend on what a task requires and how a business values extra performance against additional cost and delay.
Anthropic’s staged access to Mythos also means organizations seeking that version must meet the company’s verification requirements for cybersecurity or life sciences. The supplied material gives no further timetable or access criteria. Buyers should look for additional independent testing and clearer workload-specific results before treating the index ranking as a forecast of their own outcomes.
Key Questions
What does Claude Fable 5.2’s No. 1 ranking mean?
Artificial Analysis ranked Fable 5.2 first among 74 models on its Intelligence Index when Fable was set to Max effort, with a displayed score of 65. The ranking applies to that benchmark, not every use case.
How much did a Fable 5.2 task cost in the benchmark?
Artificial Analysis reported average API costs of $3.10 at Low, $3.80 at Medium, $9.08 at High, $14.83 at Xhigh and $26.54 at Max. These are averages for tasks in its index, not guaranteed prices for individual real-world requests.
Is Max effort always better than High?
No. Max scored 65 versus High’s 62 on the Intelligence Index, but the supplied material says Fable scored 48.8% at Max and 49.1% at High on Artificial Analysis’s FrontierCode 2.0 test. Results depend on the evaluation and task.
What is the difference between Fable 5.2 and Mythos 5.2?
The supplied source describes them as the same underlying model with different access and safety rules. Fable is available generally and has safeguards for certain high-risk cyber and biology requests; Mythos is restricted to organizations accepted into Anthropic’s verification programs.
Are Anthropic’s demonstration results independently verified?
Not according to the supplied material. The reported demonstrations, including the Fermat’s Last Theorem run, are Anthropic’s own accounts. They should not be treated as independently confirmed performance results.
Source: ThorstenMeyerAI.com
