🔍 Read the full analysis: Claude Fable 5.2 At #1: A Cost Guide For Every Effort Setting on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Artificial Analysis ranked Anthropic’s Claude Fable 5.2 first among 74 models on its Intelligence Index at Max effort, scoring 65. The same model scored 62 at High effort for $9.08 per test task, compared with $26.54 at Max; these figures are benchmarks, not a guaranteed cost for everyday use.
Anthropic released Claude Fable 5.2 on November 9, and independent model evaluator Artificial Analysis ranked it No. 1 of 74 on its Intelligence Index at the Max effort setting the next day. The top score of 65 came with an average benchmark cost of $26.54 per task; at High effort, the model scored 62 for $9.08, giving businesses a concrete trade-off between performance, cost and waiting time.
Artificial Analysis tested five effort settings on its Intelligence Index, which combines nine evaluations covering agentic coding, business documents, science and knowledge. It reported scores of 54 at Low, 56 at Medium, 62 at High, 62 at Xhigh and 65 at Max. Those results placed the settings at 15th, ninth, third, second and first, respectively, among 74 models. Scores are rounded as displayed by the evaluator.
The evaluator’s average cost per task rose from $3.10 at Low and $3.80 at Medium to $9.08 at High, $14.83 at Xhigh and $26.54 at Max. Average task time also increased, from 6.5 minutes at Low to 46.4 minutes at Max. These are costs and times for the evaluator’s benchmark tasks, not a fixed charge or service-time promise for every user request.
Anthropic lists the same standard API rates across Fable’s effort settings: $8 per million input tokens and $40 per million output tokens. The difference in benchmark bills reflects how much the model uses, including tokens spent reasoning. Artificial Analysis reported 34.9 million output tokens across its index at Low and 275 million at Max. Medium is the default in Claude apps and Claude Code, according to the source report.
Ranked No. 1 at Max What each effort setting costs
Artificial Analysis ranked Claude Fable 5.2 first among 74 models on its Intelligence Index at Max effort. High effort scored 62 for a lower benchmark cost, offering buyers a practical trade-off to test.
Effort changes the benchmark trade-off
Scores and average costs rose across settings, but these are evaluator task results—not fixed prices for everyday requests.
| Effort | Score | Rank | Avg. task cost |
|---|---|---|---|
| Low | 54 | 15th | $3.10 |
| Medium | 56 | 9th | $3.80 |
| High | 62 | 3rd | $9.08 |
| Xhigh | 62 | 2nd | $14.83 |
| Max | 65 | 1st | $26.54 |
The cost of pushing higher
Benchmark task time also increased from 6.5 minutes at Low to 46.4 minutes at Max.
What buyers should weigh
Choose settings against the needs and limits of the work, not the index rank alone.
Benchmark is not a quote
Actual API bills depend on prompts, outputs, effort settings and token use. Anthropic lists standard rates of $8 per million input tokens and $40 per million output tokens.
Results vary by task
On FrontierCode 2.0, High scored 49.1% and Max 48.8%. The top index result does not guarantee the best setting for every workload.
Separate data from claims
The ranking is Artificial Analysis benchmark data. Faster answers and fewer safety blocks are Anthropic claims, not independently verified findings in the provided material.
A practical evaluation path
Use representative work to find where added effort earns its cost and wait time.
Medium is the default in Claude apps and Claude Code; Anthropic recommends starting with Claude Opus 5.5 for most coding and knowledge work.
Compare output quality, latency and token use at relevant effort settings.
Check whether High or Max adds enough value for a specific workflow to justify added cost and time.
The Price of Higher Effort
The ranking gives buyers more than a single score: it shows how substantially the bill and response time can change when the same model is set to think harder. In Artificial Analysis’s test, moving from Medium to High added six score points for $5.28 more per task. Moving from High to Xhigh raised the reported cost to $14.83 while leaving the rounded score at 62. Max added three points over High but cost almost three times as much and took an average of 46.4 minutes.
Those comparisons can help teams decide which setting to test for work where extra capability may justify more usage and delay. They do not prove that Max is the best choice for every difficult request. On Artificial Analysis’s FrontierCode 2.0 coding test, Fable 5.2 at High scored 49.1%, slightly above its 48.8% result at Max. Benchmark results can differ by task, and actual API bills depend on the tokens used for each request.
The source report also compares Fable 5.2 at High with Anthropic’s Claude Opus 5.5 at Max: Artificial Analysis scored them at 62 and 59, with benchmark costs of $9.08 and $10.04 per task, respectively. That is a useful comparison within this particular index, but it should not be treated as a universal measure of value across workloads.
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.
Fable and Mythos Release
Anthropic released Fable 5.2 and Mythos 5.2 as replacements for their 5.1 counterparts. According to the source report, Anthropic describes them as the same underlying model with different access and safety controls: Fable is available to users generally, while Mythos is restricted to organizations accepted into its Cyber or Life Sciences Verification Programs.
Anthropic says Fable’s safeguards block some high-risk cybersecurity work, including penetration testing and exploit generation, and most tasks in research biology. The report says blocked requests receive responses from older models. Anthropic also reported that early testing showed about 55% fewer blocks than Fable 5.1. Those are company statements; the provided material does not include independent verification of the safety results.
Anthropic also claims Fable 5.2 writes answers up to 30% faster than version 5.1 and describes demanding prerelease demonstrations, including a proof-related run using about 1,000 agents. Such company-reported demonstrations are not the same as independent benchmark results or evidence of routine performance. The ranking and cost figures discussed here come from Artificial Analysis data accessed November 10, while the safety and demonstration claims come from Anthropic.
“For most coding and knowledge work, Anthropic recommends starting with Claude Opus 5.5 and moving up to Fable only when the task needs it.”
— Anthropic, as reported by ThorstenMeyerAI.com
Benchmark Limits and Open Questions
The ranking establishes Fable 5.2’s performance on Artificial Analysis’s stated index, not on every business workload. The source material does not provide the underlying results for every individual evaluation or enough detail to predict a customer’s bill from a benchmark average. Costs in practice depend on prompts, outputs, effort settings and token usage.
Some of the most striking performance examples in the report are Anthropic’s own prerelease claims, not independently verified findings. The same applies to its reported safety-block reduction and faster answer generation. It is also not clear from the provided material how access to Mythos will be approved in practice, how broadly it will be offered, or how the safety controls will perform across real-world use.
The source article discloses that it was drafted with assistance from Anthropic’s Claude, while stating that Anthropic and Artificial Analysis did not review or sponsor it. Readers should keep that disclosure, the distinction between vendor claims and evaluator data, and the test-specific nature of the cost figures in mind.
What Buyers Should Test Next
The immediate next step for organizations considering Fable is to compare effort settings on their own representative tasks, tracking output quality, latency and token use rather than relying only on index averages. A team could start with the Medium default or Anthropic’s recommended Opus 5.5 starting point, then test whether High or Max provides enough added value for a specific workflow to justify the extra cost and time.
Future updates to Artificial Analysis’s index or additional independent evaluations could show whether the ranking holds across new tests and model versions. Anthropic’s availability and verification arrangements for Mythos are also likely to matter to organizations working in cybersecurity or life sciences, but the source material does not give a timeline for changes to that access. For now, the confirmed comparison is a snapshot: Fable 5.2 led this index at Max, while High delivered a lower-cost score of 62 in the same evaluation.
Key Questions
What does “No. 1” mean in this report?
Artificial Analysis ranked Fable 5.2 at Max effort first among 74 models on its Intelligence Index, with a score of 65. The result applies to that index, not every possible task or ranking.
How much does a Fable 5.2 task cost?
In Artificial Analysis’s benchmark, average cost ranged from $3.10 at Low to $26.54 at Max. Those are evaluator-reported costs per index task, not a standard price for every API request. Anthropic’s listed standard rates are $8 per million input tokens and $40 per million output tokens.
Which effort setting offers the lower-cost alternative to Max?
High scored 62 for an average $9.08 per benchmark task, compared with 65 for $26.54 at Max. The best setting for a particular user depends on the task, quality needs and tolerance for latency.
Who can use Claude Mythos 5.2?
According to the source report, Anthropic restricts Mythos 5.2 to organizations accepted into its Cyber or Life Sciences Verification Programs. The report describes Fable 5.2 as the generally available version with stronger safeguards for certain high-risk requests.
Are Anthropic’s demonstration claims independently verified?
Not in the material provided. The reported demonstrations and safety claims are attributed to Anthropic; the index scores and task-cost estimates are attributed to Artificial Analysis.
Source: ThorstenMeyerAI.com
