🔍 Read the full analysis: Claude Fable 5.2: A Practical Look At Effort Levels And Task Pricing 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 November 9, 2026, with five effort settings that produce different benchmark scores, speeds and estimated task costs. Artificial Analysis ranked the Max setting first on its Intelligence Index, but reported a $26.54 average cost per task, compared with $9.08 at High; the benchmark results do not establish which setting is best for every business task.
Anthropic released Claude Fable 5.2 on November 9 with five effort settings, giving users a choice between lower-cost, faster responses and more model computation. On Artificial Analysis’s Intelligence Index, Max effort ranked first among 74 models with a score of 65, but its average benchmark task cost was $26.54, compared with $9.08 at High effort, which scored 62.
Artificial Analysis evaluated the model across nine assessments covering areas including agentic coding, business documents, science and knowledge. Its reported cost per task is an average API cost for a task in that index, not a fixed price for every real-world request. The results put Fable 5.2 at Low effort at a score of 54 and $3.10 per task; Medium, the default in Claude apps and Claude Code, scored 56 at $3.80. High scored 62 at $9.08, Xhigh scored 62 at $14.83, and Max scored 65 at $26.54.
The rankings illustrate that spending more does not always raise the displayed score. Xhigh cost 63% more than High while both scored 62 as rounded by Artificial Analysis. Max added three points over High, while its estimated task cost was nearly three times as much. Average task time also increased, from 17.9 minutes at High to 46.4 minutes at Max; Low averaged 6.5 minutes. On Artificial Analysis’s FrontierCode 2.0 coding test, Fable 5.2 at High scored 49.1%, slightly above Max at 48.8%.
Anthropic says the release replaces Fable 5.1 and that the new model produces answers up to 30% faster than its predecessor. The company lists API prices of $8 per million input tokens and $40 per million output tokens for each effort level. The difference in the index cost estimates reflects how much the model uses, including output tokens spent reasoning: Artificial Analysis reported 34.9 million output tokens across the index at Low and 275 million at Max. Anthropic also offers an optional fast mode, which it says generates tokens up to 2.5 times faster at double the listed per-token prices.
Claude Fable 5.2 Effort, cost and results
Five effort settings trade benchmark results against estimated task cost and time. The best fit depends on the work your team needs done.
Effort changes the trade-off
Artificial Analysis scores and costs reflect its benchmark index, not every real-world request.
| 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 |
More effort is not always better
The leading setting on one evaluation can lose on another.
- Max scored 65, the highest among 74 models.
- High scored 62 at a lower estimated task cost.
- High scored 49.1%.
- Max scored 48.8%.
Choose settings for the work
Compare quality, response time and token use on representative tasks before making a setting the default.
Use work that reflects your team’s actual needs.
Track output quality, latency and token use.
Check whether higher effort improves results enough to justify the expense.
The Cost of More Model Effort
The results make effort selection a budgeting and latency decision, not simply a setting to leave at its maximum. For organizations running many model tasks, differences in average token use can affect API bills; for users waiting on an answer, the reported gap between 17.9 and 46.4 minutes may matter as much as the cost. Those timings are averages from this benchmark and should not be treated as a forecast for every application.
The index also offers a comparison with Anthropic’s Claude Opus 5.5. Artificial Analysis lists Opus 5.5 at Max effort at a score of 59 and $10.04 per task, while Fable 5.2 at High scored 62 at $9.08. That is a benchmark comparison, not proof of lower costs or better results in a particular company’s workflow. Actual expenses and performance depend on task, prompts, usage and system setup. The figures give teams a reason to test settings on representative work before adopting Max by default.
Anthropic’s stated guidance points in a similar direction: the company recommends starting with Opus 5.5 for most coding and knowledge work, then moving to Fable when a task calls for it. This is vendor advice, not an independent finding. The practical takeaway from the published index is narrower: higher effort can buy a higher score on some evaluations, but the gains and costs vary by task.
Products Worth Considering
LG UltraGear 27-inch QHD OLED Gaming Monitor, 240Hz
LG UltraGear 25G590B-B 25-Inch Full HD 1000Hz Gaming Monitor
LG UltraGear 32GX850A-B 32-inch 4K UHD OLED Gaming Monitor
As an affiliate, we earn on qualifying purchases.
Two Names, Different Access
Anthropic describes Fable 5.2 as available to users with safeguards for high-risk cybersecurity work and much research biology. According to the company, its classifiers block requests such as penetration testing or exploit generation and most tasks in research biology; blocked requests receive answers from older models. Anthropic says early testing found about 55% fewer blocks than with Fable 5.1. That figure is the company’s report, and the source material does not provide an independent verification.
The same release includes Claude Mythos 5.2, which Anthropic describes as the same underlying model without some of Fable’s safety restrictions. The company says Mythos is limited to organizations accepted into its Cyber or Life Sciences Verification Programs. The distinction concerns access and safeguards, not separate benchmark results presented in the supplied material.
Anthropic also publicized ambitious prerelease demonstrations, including work on electronic devices and mathematical problems. The source account says the company reported that a team of about 1,000 agents formalized a proof of Fermat’s Last Theorem in 30 hours. Anthropic’s comparison of that run with Opus 5.5 is based on one run per model and is vendor data, not independently verified. Such demonstrations describe reported results under specific conditions, not typical performance for routine business use.
“The company recommends starting with Claude Opus 5.5 for most coding and knowledge work, and moving to Fable when a task needs it.”
— Anthropic
Limits of the Benchmark Evidence
The supplied figures do not show how the five settings perform across a broad range of live customer workloads. Artificial Analysis’s cost-per-task figure applies to its index evaluations; the material does not establish a typical bill for a business using Fable in production. Nor does the rounded score show whether small differences between effort levels would persist across repeated runs.
Anthropic’s speed, safety and prerelease achievement claims are company statements. The source material does not include independent verification of those claims, the full test methods for every demonstration, or evidence that benchmark differences will translate into specific productivity gains. It also does not give enough detail to determine how the models’ safeguards behave across all possible requests or how Mythos access decisions are made.
One result cautions against treating effort as a simple quality scale: High slightly outscored Max on FrontierCode 2.0, while the Intelligence Index gave Max a higher overall score. Performance is evaluation-dependent, and teams will need to check their own task mix, response-time requirements and costs.
Test Settings Against Real Work
The immediate next step for teams considering Fable 5.2 is to compare settings on a sample of their own tasks, tracking output quality, latency and token use rather than relying on a single overall index score. A staged test can show whether the additional cost of High or Max changes results enough to justify it, and whether some work is better handled by a lower setting or Opus 5.5.
Readers should also watch for further independent evaluations and more detailed reporting on real-world usage. The source material gives no schedule for new benchmark releases or changes to Mythos access. For now, the confirmed public comparison is the November 10 snapshot: Max leads the index ranking, while lower effort levels cost less and can be faster, with results that vary by evaluation.
Key Questions
What is Claude Fable 5.2?
Claude Fable 5.2 is an Anthropic model released November 9, 2026, with selectable effort levels from Low to Max. Anthropic says it replaces Fable 5.1 and includes safeguards for some high-risk cybersecurity and biology requests.
How much does each effort level cost in the benchmark?
Artificial Analysis reported average costs 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 benchmark estimates, not guaranteed prices for an individual user request.
Does Max always perform better than High?
No. Max scored higher on the Intelligence Index, at 65 versus 62 for High, but both settings scored 62 on the rounded index result at Xhigh and High. On FrontierCode 2.0, High scored 49.1% and Max 48.8%, according to the supplied source.
What is Claude Mythos 5.2?
Anthropic describes Mythos 5.2 as the same underlying model as Fable 5.2 without some of its safety restrictions. The company says access is limited to organizations accepted into its Cyber or Life Sciences Verification Programs.
Are the reported demonstrations independently verified?
The source material attributes the demonstrations and their results to Anthropic. It describes the Fermat’s Last Theorem comparison as one run per model and says the figures are vendor data, not independently verified.
Source: ThorstenMeyerAI.com
