🔍 Read the full analysis: Claude Fable 5.2 Pricing By Effort Level: What Each Task Costs 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
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, while scores range from 54 to 65. The figures are test results, not fixed prices for every real-world task; actual API charges depend on token use. Anthropic released Fable 5.2 on November 9, 2026, alongside the access-restricted Mythos 5.2.
Anthropic released Claude Fable 5.2 on November 9, and testing by Artificial Analysis shows that its average cost per Intelligence Index task ranges from $3.10 at Low effort to $26.54 at Max. The comparison gives businesses a way to weigh test performance against expense and response time, but the published task costs are averages from a specific benchmark, not a price quote for every use.
Artificial Analysis tested five settings: Low, Medium, High, Xhigh and Max. It reported scores of 54, 56, 62, 62 and 65, respectively, on its Intelligence Index. The corresponding average cost per task was $3.10, $3.80, $9.08, $14.83 and $26.54. The test covers nine evaluations spanning areas including agentic coding, business documents, science and knowledge work. Scores are rounded as displayed by Artificial Analysis.
The benchmark’s reported average time per task rose from 6.5 minutes at Low to 46.4 minutes at Max. Artificial Analysis also reported that total output tokens used across the index grew from 34.9 million at Low to 275 million at Max. These are aggregate test measurements, not token totals for a single task. Medium is the default effort setting in Claude apps and Claude Code, according to the supplied source.
Anthropic lists standard API rates of $8 per million input tokens and $40 per million output tokens for Fable 5.2 at every effort setting. The effort dial changes how much the model thinks and, consequently, how many tokens it uses. Anthropic also offers an optional fast mode at $16 per million input tokens and $80 per million output tokens; the company says it can generate tokens up to 2.5 times faster.
Claude Fable 5.2 Pricing By Effort Level
Artificial Analysis found a wide cost range across five effort settings. These are benchmark averages, not fixed prices for every real-world task.
Cost and score by effort
The benchmark’s score rose overall with effort, while average task costs increased substantially.
| Effort | Score | Average cost |
|---|---|---|
| Low | 54 | $3.10 |
| Medium | 56 | $3.80 |
| High | 62 | $9.08 |
| Xhigh | 62 | $14.83 |
| Max | 65 | $26.54 |
What the higher settings change
- High scored six points above Medium.
- Max recorded the highest score in the test.
- Higher effort may suit difficult work where added performance matters.
- High cost $5.28 more per task than Medium on average.
- Xhigh cost more than High while both displayed a score of 62.
- Max took longer on average than Low.
API pricing stays fixed; usage varies
Anthropic lists the same standard token rates at every effort level. Effort can change token use, which changes the resulting bill.
$8 per million tokens
Standard API input rate for Fable 5.2.
$40 per million tokens
Standard API output rate for Fable 5.2.
$16 input · $80 output
Optional rates; Anthropic says it can generate tokens up to 2.5 times faster.
Test your own workload
Benchmark averages cannot predict costs or results for every company’s prompts and workflows.
Use work your team actually handles.
Record quality, token use, cost and completion time.
Balance the value of performance against expense and response time.
How the Effort Setting Changes Cost
The figures make effort selection a budget and latency decision, not simply a choice of model quality. In this benchmark, moving from Medium to High raised the score by six points, from 56 to 62, while increasing average task cost by $5.28. The next step, from High to Xhigh, cost more but kept the displayed score at 62. That does not establish that Xhigh offers no benefit on other tasks; it means the benchmark’s rounded score did not rise.
At Max, the score reached 65, three points above High, while the reported task cost was nearly three times as high. Max also took substantially longer on average. For teams handling large numbers of requests or work that needs a quick response, those differences may matter as much as the peak score. Conversely, difficult work where a small performance gain is valuable may justify a higher setting. The test does not show which level is best for every business workflow.
Performance did not rise on every evaluation as effort increased. On Artificial Analysis’s FrontierCode 2.0 agentic coding test, the supplied source reports 49.1% at High and 48.8% at Max. That result is a reminder that effort settings should be evaluated against the work a team actually does, rather than assumed to deliver a steady improvement across all tasks.
Products Worth Considering
FLEXISPOT Electric Adjustable Workbench, 55 x 28 Inches, White
Apple Mac mini Desktop Computer with M2 Pro, 16GB Memory, 512GB SSD, Gigabit Ethernet
Logitech MX Master 4 Ergonomic Wireless Mouse with Haptic Feedback, Graphite
As an affiliate, we earn on qualifying purchases.
Fable, Mythos and Anthropic’s Release
Anthropic announced Fable 5.2 and Mythos 5.2 as replacements for their 5.1 versions. The source describes Mythos as the same underlying model with fewer safety restrictions, available only to organizations accepted into Anthropic’s Cyber or Life Sciences Verification Programs. Fable is the generally available version, with safeguards for high-risk cybersecurity work and many research-biology tasks. Anthropic says requests blocked by those safeguards receive responses from older models.
Anthropic says Fable 5.2 writes answers up to 30% faster than its 5.1 version and that early testing showed about 55% fewer blocked requests. Those are company claims, not findings from the Artificial Analysis cost table. Anthropic also described striking prerelease demonstrations involving electronics, physics and a multi-agent attempt to formalize Fermat’s Last Theorem. The supplied material attributes those accounts to Anthropic; they are not independent verification of typical performance.
Artificial Analysis placed Fable 5.2 at the top of its 74-model Intelligence Index at Max effort, with High and Xhigh also among its top-ranked entries. The ranking and cost figures reflect the organization’s specified benchmark and test conditions. They should not be treated as a universal measure of performance or a guarantee of results on a particular customer’s tasks.
“Start lower.”
— Anthropic
What Benchmark Costs Cannot Show
The reported cost per task is Artificial Analysis’s average API cost for a task in its Intelligence Index. The source does not provide enough detail here to calculate what a company would spend on its own prompts, agent setups or workload mix. Real charges can vary with input and output token use, and the benchmark averages should not be read as fixed prices for individual requests.
The supplied material does not include the complete test methodology, uncertainty ranges or results for every business use case. The score differences are rounded, and a tie at 62 does not prove the settings are equivalent. Nor does a high index ranking guarantee that Fable 5.2 will outperform another model on a particular job.
Anthropic’s safety, speed and prerelease capability statements remain company claims in the material provided. The account does not establish how often safeguards will block real requests, or how the model’s results will hold up across independent, production-scale deployments. The article’s source also discloses that its analysis was drafted with help from Claude, an Anthropic model, while stating that Anthropic and Artificial Analysis did not review or sponsor it.
Testing Against Real Workloads
For teams considering Fable 5.2, the practical next step is to compare settings on representative tasks while recording quality, token use, cost and completion time. A benchmark-based starting point could be Medium for routine work and High for tasks where the observed score gain matters, but that is an evaluation approach rather than a universal recommendation. Anthropic’s stated advice is to start with Opus 5.5 for most coding and knowledge work and move up when needed.
Further independent testing and production experience will help clarify how the benchmark differences translate to actual workloads. The source provides no announced date for another evaluation or for changes to model pricing. Until then, the listed figures are a snapshot of Artificial Analysis results accessed on November 10, 2026, alongside Anthropic’s published rates.
Key Questions
How much does a Fable 5.2 task cost?
Artificial Analysis reported an average of $3.10 per task at Low effort, rising through $3.80 at Medium, $9.08 at High and $14.83 at Xhigh to $26.54 at Max. These are averages for tasks in its Intelligence Index, not fixed charges for every request.
What does the effort level change?
The setting controls how much effort the model spends on a response. Anthropic’s listed standard token rates stay the same across settings, but token use can rise with effort, changing the resulting bill. In the index test, reported output-token use increased with the effort level.
Which effort level scored highest?
Max scored 65 on the Artificial Analysis Intelligence Index, the highest reported result among the five settings. It also had the highest average task cost and longest average completion time in the supplied comparison.
Is Mythos 5.2 available to everyone?
No. According to the supplied source, Mythos 5.2 is restricted to organizations accepted into Anthropic’s Cyber or Life Sciences Verification Programs. Fable 5.2 is the version described as open to all users, with safeguards for certain high-risk requests.
Are the reported task costs guaranteed for my work?
No. They are benchmark averages from Artificial Analysis. A real task’s API cost depends on its token use and setup, so teams would need to measure their own workloads to estimate spending.
Source: ThorstenMeyerAI.com
