🔍 Read the full analysis: Claude Fable 5.2 Is The New #1 AI Model. What Each Effort Level Costs You Per Task 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 Nov. 9, and Artificial Analysis ranked it first among 74 models at Max effort the following day. The same test shows a wide cost range: $3.10 per task at Low effort and $26.54 at Max, with High scoring 62 for $9.08. Anthropic’s separate claims about research achievements have not been independently verified.
Anthropic released Claude Fable 5.2 on Nov. 9, and a day later the model ranked No. 1 among 74 models on Artificial Analysis’s Intelligence Index at its highest effort setting. The ranking comes with a significant cost difference across settings: the index’s estimated cost per task ranged from $3.10 at Low to $26.54 at Max, while the Max score was 65 and the Low score was 54.
Artificial Analysis tested Fable 5.2 at five effort levels on a set of nine evaluations covering agentic coding, business documents, science and knowledge. It reported a score of 54 and a $3.10 average task cost at Low; Medium, the default in Claude apps and Claude Code, scored 56 for $3.80. High scored 62 for $9.08, Xhigh scored 62 for $14.83, and Max scored 65 for $26.54. The costs are averages for tasks in that index, not a fixed fee for every real-world request.
The results suggest the performance and spending curve is uneven. Moving from Medium to High added six score points for $5.28 more per task in the test. Xhigh cost more than High while retaining the same rounded score; Max added three points over High and cost nearly three times as much. Average task time also rose from 6.5 minutes at Low to 46.4 minutes at Max. On Artificial Analysis’s FrontierCode 2.0 coding test, Fable scored 49.1% at High and 48.8% at Max, so the highest setting did not lead on that test.
Anthropic lists the standard API price as $8 per million input tokens and $40 per million output tokens at every effort level. The different estimated task bills reflect how many tokens the model uses, including tokens spent reasoning: Artificial Analysis reported 34.9 million output tokens across the full index at Low and 275 million at Max. Anthropic also offers an optional fast mode at $16 per million input tokens and $80 per million output tokens, which it says can generate tokens up to 2.5 times faster.
Claude Fable 5.2 leads the index What each effort level costs
Artificial Analysis ranked Fable 5.2 first among 74 models at Max effort. Its test shows higher effort can mean higher costs and longer waits.
Cost rises unevenly with effort
Artificial Analysis estimated average costs per task across five effort settings.
The score gain depends on the task
- High scored 62 for $9.08 per task.
- Max scored 65 for $26.54 per task.
- High scored 49.1% on FrontierCode 2.0.
- Max scored 48.8% on the same test.
What the test means for work
The highest overall score does not guarantee the best result for every workload.
Average task time rose from 6.5 minutes at Low to 46.4 minutes at Max.
Measure quality, latency and token costs before choosing a setting for deployment.
Access and claims
Available with safeguards
The source describes safeguards for high-risk cyber work and most research biology tasks.
Restricted access
Organizations must be accepted into Anthropic’s Cyber or Life Sciences Verification Programs.
The Cost of More Model Effort
The ranking does not mean every user should select Max. For businesses, the practical issue is whether extra performance on a particular task justifies higher API costs and longer waits. In this index, High scored 62 for $9.08 per task, while Max scored 65 for $26.54. That makes the setting choice relevant to both budget planning and workflows where people are waiting for results.
The numbers also caution against treating one overall benchmark position as a complete guide to deployment. Results vary by evaluation: High slightly outscored Max on the cited coding test, despite Max leading the combined index. Artificial Analysis’s measurements offer a comparison under its test conditions; they do not establish the best setting for every company’s software, prompts or workload. Anthropic’s recommendation to start with its Opus model for most coding and knowledge work, and step up to Fable when needed, is the company’s guidance rather than an independent finding.
Products Worth Considering
LG UltraGear 25G590B-B 25-Inch Full HD 1000Hz Gaming Monitor
Samsung Galaxy Tab A11+ 8GB RAM, 256GB Storage, Gray
Samsung Galaxy Tab A11+ 11-Inch Tablet, 6GB RAM, 128GB Storage, Expandable, Gray
As an affiliate, we earn on qualifying purchases.
Two Names, Different Safety Access
Anthropic introduced Fable 5.2 and Mythos 5.2 as replacements for the 5.1 versions. The source material describes them as the same underlying model, with a different access and safety arrangement: Fable is available to users with safeguards for high-risk cyber work and most research biology tasks, while Mythos is reserved for organizations accepted into Anthropic’s Cyber or Life Sciences Verification Programs. Anthropic says blocked Fable requests are answered by older models instead.
Anthropic says the new models produce answers up to 30% faster than their 5.1 counterparts and reports that early testing found about 55% fewer blocks with Fable 5.2 than with Fable 5.1. Those are company-reported figures. The source also describes Anthropic’s claims about prerelease demonstrations, including a multi-agent effort to formalize Fermat’s Last Theorem; these results are not independently verified and should not be confused with the Artificial Analysis ranking.
The article source discloses that its analysis was drafted with help from Claude, an Anthropic model, and says neither Anthropic nor Artificial Analysis reviewed or sponsored it. That disclosure is relevant when weighing the source’s discussion of the company’s product and claims.
“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 Results
The source gives Artificial Analysis scores and estimated costs but does not provide enough detail here to establish how closely its tasks match any particular company’s workload. Its per-task figures are averages from the index, and actual API bills will depend on usage, task length and model behavior. Scores are reported as rounded values, so equal scores for High and Xhigh do not prove the settings perform identically in every evaluation.
Anthropic’s claims about speed, fewer safety blocks and research demonstrations remain company claims in the supplied material. It is not clear from that material how the 55% reduction in blocks was measured, or whether independent parties have reproduced the reported scientific and engineering results. The source also does not specify when the verification programs may accept additional organizations or how access decisions are made.
Choosing a Setting for Real Work
For users deciding whether to adopt Fable 5.2, the immediate next step is to compare effort settings on their own tasks and track quality, latency and token costs, rather than assume the highest benchmark score is always worth the added expense. Businesses considering Mythos will need to check Anthropic’s verification-program requirements. Further independent testing could clarify how the model’s benchmark gains translate to day-to-day work and whether the company’s separate performance and safety claims hold across broader use.
Key Questions
What does the No. 1 ranking refer to?
Artificial Analysis ranked Fable 5.2 at Max effort first among 74 models on its Intelligence Index, with a score of 65. The ranking applies to that index and its evaluation set, not every possible AI task.
How much does each effort level cost in the test?
Artificial Analysis estimated an average cost per index task of $3.10 at Low, $3.80 at Medium, $9.08 at High, $14.83 at Xhigh and $26.54 at Max. These are test averages, not guaranteed prices for individual requests.
Is Max effort always the best choice?
No. Max earned the highest overall index score, but it took an average of 46.4 minutes per task and cost nearly three times as much as High in the test. Fable also scored slightly higher at High than Max on the cited FrontierCode 2.0 evaluation.
Who can use Mythos 5.2?
According to the source material, Mythos is restricted to organizations accepted into Anthropic’s Cyber or Life Sciences Verification Programs. Fable is the version offered with safeguards for high-risk cyber and biology requests.
Source: ThorstenMeyerAI.com
