🔍 Read the full analysis: The Budget Question For Claude Fable 5.2: Cost By Effort Level on ThorstenMeyerAI.com
Get hardware and tech essentials 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
Artificial Analysis testing puts Claude Fable 5.2’s Intelligence Index score at 54 on Low effort and 65 on Max, while the estimated cost per test task rises from $3.10 to $26.54. High scored 62 at $9.08; Xhigh also scored 62 on the rounded scale at $14.83. The figures illustrate a trade-off between performance, cost and speed, not a guarantee of results on every business task.
Claude Fable 5.2’s cost varies substantially with its effort setting, according to Artificial Analysis testing published November 10: estimated cost per Intelligence Index task ranged from $3.10 at Low to $26.54 at Max. The Max setting scored 65, compared with 54 at Low, giving businesses a measurable price-and-performance trade-off rather than one fixed cost for using the model.
Artificial Analysis tested Fable 5.2 at five settings: Low, Medium, High, Xhigh and Max. Its reported Intelligence Index scores were 54, 56, 62, 62 and 65, respectively. The corresponding average API costs per task were $3.10, $3.80, $9.08, $14.83 and $26.54. The index covers nine evaluations in areas including agentic coding, business documents, science and knowledge work. The cost figure is an average for a task in that test, not a quoted price for every request in ordinary use.
The reported average time per task also climbed with effort: 6.5 minutes at Low, 8.3 at Medium, 17.9 at High, 26.4 at Xhigh and 46.4 at Max. Artificial Analysis says the model has the same listed token rates at each setting: $8 per million input tokens and $40 per million output tokens. The difference in the tested bill comes from how many tokens the model uses while working. Across the full index, reported output-token use rose from 34.9 million at Low to 275 million at Max.
The largest score jump in the reported figures was between Medium and High: six points for $5.28 more per task. Xhigh cost about 63% more than High but had the same rounded score, 62. Max added three points over High while costing almost three times as much. Scores are shown as Artificial Analysis reports them, rounded; the figures do not establish that every individual task benefits from a higher setting.
The budget question cost by effort level
Artificial Analysis testing shows higher effort can raise benchmark scores, but costs and task times rise sharply too. Treat these figures as test averages, not a forecast of your business bill.
Five effort settings, five cost profiles
Artificial Analysis reported these average costs and Intelligence Index scores per test task.
| Effort | Score | Cost per task | Time per task |
|---|---|---|---|
| Low | 54 | $3.10 | 6.5 min |
| Medium | 56 | $3.80 | 8.3 min |
| High | 62 | $9.08 | 17.9 min |
| Xhigh | 62 | $14.83 | 26.4 min |
| Max | 65 | $26.54 | 46.4 min |
Where the trade-off sharpens
The upper settings cost more for smaller score gains in the reported index.
Higher cost, same rounded score
Xhigh costs about 63% more than High, while both score 62 on the rounded index.
Three more points
Max scores 65, but costs almost three times as much as High and takes longer per test task.
Largest score jump
The score rises by six points for $5.28 more per task in the benchmark.
What drives the tested bill?
The listed per-token rates are the same across settings; the model uses more tokens as effort increases.
- $8 per million input tokens
- $40 per million output tokens
- Output-token use rises from 34.9 million at Low to 275 million at Max across the index
- Average time per task also increases
Use benchmark results with care
A higher setting is not a guaranteed improvement on every task.
- High and Xhigh both report a rounded score of 62; rounding does not prove identical underlying results.
- On Artificial Analysis’s FrontierCode 2.0 test, High scored 49.1% and Max scored 48.8%.
- Real costs depend on your prompts, output length, task mix and volume.
- The benchmark covers nine evaluations and may not match a typical commercial workload.
A practical buying test
Measure the settings on representative work before choosing a default.
Use prompts and workloads that reflect your team’s needs.
Record cost per completed task, turnaround time and whether quality requirements are met.
Reserve higher settings for work where your evaluation shows the added effort is worthwhile.
The Price of More Model Effort
The figures matter for organizations budgeting model use because the effort setting can change both API spending and response time. A team processing many tasks could face meaningfully different costs depending on whether it defaults to Medium, uses High for demanding work, or sends requests to Max. The test suggests that choosing the highest setting by default may be difficult to justify when the score gain is small relative to the added cost and delay.
But the index is a benchmark, not a forecast of a company’s monthly bill or proof that a particular setting is best for its workflow. Real costs depend on task mix and token use, while a benchmark score compresses performance across a fixed set of evaluations. A slower setting may still be worthwhile for a task where accuracy matters more than turnaround time; a cheaper setting may suffice for routine work. Businesses would need to test their own prompts and track quality, latency and spending before making that choice.
Artificial Analysis also reported that Fable 5.2 at High scored 49.1% on its FrontierCode 2.0 agentic coding test, slightly above the 48.8% result at Max. That result cautions against treating effort as a simple quality dial that always moves in one direction. It is specific to that test and does not show that High will outperform Max across other coding tasks.
Products Worth Considering
LG UltraGear 34GX90SA-W 34-inch WQHD Curved OLED Gaming Monitor
LG UltraGear 32GS60QC-B 32-inch Curved QHD Gaming Monitor (180Hz, 1ms, 1000R, AMD FreeSync, HDR10)
LG UltraGear 32GX850A-B 32-inch 4K UHD OLED Gaming Monitor
As an affiliate, we earn on qualifying purchases.
How Fable’s Effort Settings Work
Anthropic released Claude Fable 5.2 on November 9, 2026, replacing Fable 5.1, according to the source material. The model offers five effort levels, with Medium the default in Claude apps and Claude Code. Anthropic says the new model produces answers up to 30% faster than its 5.1 version, and lists API prices of $8 per million input tokens and $40 per million output tokens. Those are vendor statements; the Artificial Analysis cost-per-task figures are based on its own benchmark testing.
The source also describes Claude Mythos 5.2 as an access-restricted version of the same underlying model, available to organizations accepted into Anthropic’s Cyber or Life Sciences Verification Programs. Anthropic says public Fable includes safeguards that block certain high-risk cybersecurity requests and most research-biology tasks, while Mythos does not have those blocks. These release details help explain the product lineup, but they do not change the central budget finding: effort level drives the tested task cost even when the listed per-token rates remain the same.
Artificial Analysis ranked Fable 5.2 at Max first among 74 models on its Intelligence Index, with the Xhigh and High settings also near the top. Rank and score reflect the stated benchmark, not a universal measure of usefulness. The source material says Anthropic recommends starting with Claude Opus 5.5 for most coding and knowledge work and moving to Fable when tasks require it. That is Anthropic’s guidance, rather than an independent finding about which model suits each organization.
Limits of the Benchmark Cost Figures
The reported costs are benchmark averages, and the source does not provide a company-specific estimate or a detailed breakdown of how each real-world workload would map to the nine evaluations. Actual bills will depend on prompts, output length, chosen effort setting and task volume. It is also not clear from the supplied material how closely the benchmark’s average task mix matches typical commercial use.
The scores are rounded, so High and Xhigh both appearing at 62 does not establish that their underlying results are identical. Nor does the index show that Max’s three-point advantage over High will occur on every task. The source describes the Artificial Analysis ranking and prices as current on November 10, 2026; future pricing, benchmark results and model availability may change.
Anthropic’s release claims about speed and model capabilities are not the same as independent validation. The source notes that some dramatic capability examples came from Anthropic’s own prerelease testing. Those examples should not be treated as typical performance or as evidence that an ordinary user will achieve comparable results. The source analysis also discloses that it was drafted with help from Anthropic’s Claude, while saying neither Anthropic nor Artificial Analysis reviewed or sponsored it.
What Buyers Should Measure Next
The immediate next step for teams considering Fable 5.2 is to compare effort levels on their own representative tasks. A useful evaluation would record cost per completed task, turnaround time and whether the output meets the team’s quality requirements, rather than relying on a single aggregate score. Teams can then decide whether to reserve High or Max for work where added reasoning is demonstrably useful.
Organizations should also check Anthropic’s current documentation for pricing, model access and safety limits before deployment, since the figures here are tied to the information available on November 10, 2026. Artificial Analysis may publish updated model pages and benchmark results, but the source material does not give a schedule for future testing. Until comparable follow-up data or customer-specific evaluations are available, the cost ladder is best read as a benchmark-based guide—not a prediction of what any one deployment will cost.
Key Questions
How much does Claude Fable 5.2 cost per tested task?
Artificial Analysis reported average costs ranging from $3.10 at Low to $26.54 at Max on its Intelligence Index. These are benchmark task estimates, not a universal charge for every real-world request.
Which effort level had the best balance of score and cost?
The data does not establish one best setting for every user. High scored 62 for $9.08 per task, while Max scored 65 for $26.54. The right choice depends on a team’s task requirements, quality results and budget.
Why do task costs differ if token prices are the same?
According to the source material, all five settings use the same listed rates—$8 per million input tokens and $40 per million output tokens. Higher effort used more output tokens in the benchmark, increasing the estimated cost per task.
Did Max outperform High on every test?
No. Max scored higher on the overall Intelligence Index, but the source reports that Fable 5.2 scored 48.8% at Max and 49.1% at High on Artificial Analysis’s FrontierCode 2.0 test. Results vary by evaluation.
Are the benchmark costs a prediction of a business’s bill?
No. They are average API costs for tasks in Artificial Analysis’s test set. A business’s actual spending will depend on its workload, token use, task volume and chosen effort setting.
Source: ThorstenMeyerAI.com
