TL;DR
Get everyday essentials delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
Artificial Analysis ranked Claude Fable 5.2 first among 74 models on its Intelligence Index at Max effort, with a score of 65 and an average task cost of $26.54. Lower effort levels cost less: High scored 62 at $9.08 per task, while Low scored 54 at $3.10. The figures come from one benchmark and do not establish which setting is best for a particular business workload.
Claude Fable 5.2 took the top spot among 74 models on the Artificial Analysis Intelligence Index at Max effort, scoring 65 at an average cost of $26.54 per task. The November 10 ranking followed Anthropic’s release of the model on November 9 and highlights a trade-off for business users: lower effort settings cost less, while the benchmark score varies by setting, as explored in the earlier Fable 5.1 cost analysis.
Artificial Analysis tested five effort settings, from Low to Max, across nine evaluations covering areas including agentic coding, business documents, science and knowledge. It defines the listed cost per task as the average API cost of one task in that test. At Low, Fable scored 54 and cost $3.10 per task, averaging 6.5 minutes. The app default, Medium, scored 56 at $3.80 and 8.3 minutes.
At High, the model scored 62 at $9.08 per task and averaged 17.9 minutes. Xhigh also scored 62, at $14.83 and 26.4 minutes. Max reached 65, taking an average of 46.4 minutes per task. Artificial Analysis reports these scores rounded as displayed; the figures describe its test results, not guaranteed performance or costs for other work.
Anthropic lists the same token prices across Fable’s effort settings: $8 per million input tokens and $40 per million output tokens. The amount of output used for the benchmark’s full index run rose from 34.9 million tokens at Low to 275 million at Max. Anthropic says the new models replace Fable 5.1 and Mythos 5.1 and produce answers up to 30% faster than those predecessors. An optional fast mode costs $16 per million input tokens and $80 per million output tokens, according to the company.
Claude Fable 5.2 Takes the Top Spot Cost by Effort Level
Fable 5.2 ranked first on one AI benchmark at Max effort. Higher scores came with higher task costs and longer runtimes in the test.
Effort changes cost and score
Artificial Analysis tested five settings across nine evaluations; task costs and times are averages in that test.
| Effort | Score | Cost per task | Average time |
|---|---|---|---|
| Low | 54 | $3.10 | 6.5 minutes |
| Medium | 56 | $3.80 | 8.3 minutes |
| High | 62 | $9.08 | 17.9 minutes |
| Xhigh | 62 | $14.83 | 26.4 minutes |
| Max | 65 | $26.54 | 46.4 minutes |
What the trade-off means
Start with the task
Low and Medium cost less in the benchmark and may suit routine requests.
Pay for measured gains
Max scored three points above High, but cost nearly three times as much and took more than twice as long.
More effort is not always better
On FrontierCode 2.0, High scored 49.1% and Max 48.8%.
Access depends on the model
- Released for general users.
- Safeguards block high-risk cybersecurity requests and most research biology tasks, according to Anthropic.
- Available to organizations accepted into Anthropic’s Cyber or Life Sciences Verification Programs.
- Does not use the same blocks, according to Anthropic.
Compare on your own workload
Benchmark results may not predict performance or cost for a company’s tasks, prompts, and workflows.
Test requests your team actually handles.
Track output quality alongside cost and time.
Use higher effort where measured gains justify the added cost and wait.
The Cost of Maximum Effort
The ranking gives organizations a concrete comparison between benchmark performance and usage costs. In this test, moving from Medium to High added six score points for $5.28 more per task. Xhigh cost about 63% more than High while retaining the same rounded score of 62. Max added three points over High but cost nearly three times as much and took more than twice as long on average.
Those differences matter when teams decide how to route work through a model. A higher setting may suit tasks where a benchmark-like performance gain justifies extra cost and wait time; routine requests may not need it. These are implications of the reported test results, not a recommendation for every deployment. On Artificial Analysis’s FrontierCode 2.0 agentic coding test, Fable scored 49.1% at High and 48.8% at Max, another indication that higher effort did not improve every measured result.
Products Worth Considering
LG UltraGear 32GX850A-B 32-inch 4K UHD OLED Gaming Monitor
LG UltraGear 25G590B-B 25-Inch Full HD 1000Hz Gaming Monitor
Samsung Galaxy Tab A11+ 11-Inch Tablet, 6GB RAM, 128GB Storage, Expandable, Gray
As an affiliate, we earn on qualifying purchases.
Two Models, Different Access
Anthropic released Fable 5.2 for general users and Mythos 5.2 for organizations accepted into its Cyber or Life Sciences Verification Programs. The source describes them as sharing the same underlying model, with the distinction centered on safeguards and access. Anthropic says Fable’s classifiers block high-risk cybersecurity requests, such as exploit generation, and most tasks in research biology; blocked requests receive answers from an older model. Mythos does not use those blocks and is limited to vetted organizations working in the dual-use fields.
Anthropic says early testing found about 55% fewer blocks with Fable 5.2 than with Fable 5.1. That is the company’s reported result, and the source does not provide the test methodology. Anthropic also describes several prerelease achievements, including designing electronic devices and formalizing a proof with a large team of agents. Those accounts are company claims, not independent findings about normal customer use.
For comparison with its own models, Artificial Analysis rated Claude Opus 5.5 at Max effort 59 points and $10.04 per task. Fable 5.2 at High scored 62 for $9.08. Anthropic separately says Fable 5.2 at High scored 45.3% on FrontierCode 2.0, close to Opus 5.5’s reported best of 45.5%, at roughly half the cost per task. The two comparisons use different measures and should not be treated as interchangeable.
““For roughly half the cost per task.””
— Anthropic
Benchmark Limits and Open Questions
The ranking reflects Artificial Analysis’s nine evaluations, and does not show how Fable will perform across every company’s tasks, prompts, or workflows. The source does not provide the full benchmark methodology or enough detail to independently assess how representative the cost averages are. Actual API bills depend on usage, including the tokens generated for a given task.
It is also unclear from the source how broadly organizations will qualify for Mythos access, how the reported safety blocks were measured, or whether the benchmark ordering will change as other models are updated. The scores at High and Xhigh are both shown as 62 after rounding, so the published figures do not establish whether their underlying scores are identical.
Choosing an Effort Setting
Businesses evaluating Fable can compare its effort settings against their own tasks, tracking both output quality and the cost and time required. The benchmark suggests High may be a less expensive starting point than Max for teams seeking a score near the top of this test, but their results may differ on real workloads.
Further comparisons may clarify how the new models perform across independent evaluations and everyday use. Anthropic’s access rules for Mythos and the methodology behind its safety and speed claims may also shape how organizations assess the release. The source provides no date for a next benchmark update or broader Mythos availability.
Key Questions
What does the number-one ranking mean?
Artificial Analysis placed Fable 5.2 at Max effort first among 74 models on its Intelligence Index, with a score of 65. It is a result on that index, not a claim that the model is best for every use.
How much does Fable 5.2 cost per task?
In Artificial Analysis’s test, average task costs ranged from $3.10 at Low effort to $26.54 at Max. The test’s cost figures are averages for its index tasks; actual API costs vary with token use.
Which effort setting is the default?
Medium is the default in Claude apps and Claude Code, according to the source. Artificial Analysis measured it at a score of 56 and an average cost of $3.80 per task.
How are Fable 5.2 and Mythos 5.2 different?
Anthropic makes Fable available to general users with safeguards for high-risk cybersecurity and biology requests. Mythos is restricted to organizations accepted into Anthropic’s verification programs for those fields and does not have the same blocks, according to the source.
Does Max always perform better than High?
No. Max scored higher than High on the Intelligence Index, but both settings scored 62 on FrontierCode 2.0 as reported in the source, with High at 49.1% and Max at 48.8%. Results depend on the test.
Source: Thorsten Meyer AI
