🔍 Read the full analysis: Claude Fable 5.2 Effort Levels, Explained: What You Pay For Each Task on ThorstenMeyerAI.com
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
TL;DR
Anthropic released Claude Fable 5.2 on November 9, 2026, with five effort settings that trade task cost and speed for benchmark performance. Artificial Analysis measured average costs from $3.10 per task at Low to $26.54 at Max; those figures come from its Intelligence Index tests, not a fixed price for every real-world request.
Anthropic released Claude Fable 5.2 on November 9 with five effort settings, and new Artificial Analysis tests show that the model’s average cost per index task ranges from $3.10 at Low to $26.54 at Max. The scores rose from 54 to 65 across those settings, giving businesses a measurable trade-off between benchmark performance, latency and spending rather than one cost or score for the model as a whole.
Artificial Analysis tested Fable 5.2 on its Intelligence Index, which covers nine evaluations in areas including agentic coding, business documents, science and knowledge. Its reported cost per task is the average API cost for one task in that test. Low scored 54 and cost $3.10 per task, while the default Medium setting scored 56 at $3.80. High scored 62 at $9.08, Xhigh also scored 62 at $14.83, and Max scored 65 at $26.54. The rankings were, respectively, 15th, ninth, third, second and first among 74 models.
The reported costs do not reflect different token prices for each effort setting. Anthropic lists Fable 5.2 API pricing at $8 per million input tokens and $40 per million output tokens for every level. The difference in observed task costs reflects how much the model uses. Across the full index, Artificial Analysis measured 34.9 million output tokens at Low and 275 million at Max. The average task time also increased, from 6.5 minutes at Low to 46.4 minutes at Max.
The benchmark suggests that spending more does not guarantee a better result on every test. High and Xhigh both received a rounded score of 62, although Xhigh cost about 63% more per task. On Artificial Analysis’s FrontierCode 2.0 coding evaluation, Fable scored 49.1% at High and 48.8% at Max. These are results from specified tests, not a promise about an individual company’s workload.
Effort levels, explained What you pay for each task
Five settings trade benchmark performance for cost and time. Artificial Analysis’s figures are averages from its Intelligence Index—not fixed prices for every request.
Cost rises with effort
Average cost per Intelligence Index task increases across the five settings.
| Effort | Score | Average task cost |
|---|---|---|
| Low | 54 | $3.10 |
| Medium | 56 | $3.80 |
| High | 62 | $9.08 |
| Xhigh | 62 | $14.83 |
| Max | 65 | $26.54 |
What drives the bill
Anthropic lists the same standard token prices at every effort level; higher observed costs reflect greater usage in the test.
- $8 per million input tokens
- $40 per million output tokens
- 34.9 million output tokens at Low
- 275 million output tokens at Max
More effort is not always better
Results differ by evaluation, and spending more does not guarantee a stronger result on every task.
High vs. Xhigh
Both rounded to 62, while Xhigh cost about 63% more per task.
FrontierCode 2.0
High scored 49.1%; Max scored 48.8%.
Not a workload forecast
Index averages do not predict an individual company’s cost or results.
Choose a setting for the work
Compare quality, latency and token costs on representative tasks before scaling use.
Use representative prompts from your team’s workload.
Track answer quality, waiting time and token use.
Reserve higher-cost settings for tasks where added performance is worthwhile.
Where Higher Effort Adds Value
For teams choosing an AI setting, the results put a price on an otherwise opaque trade-off: more model effort can increase both token use and waiting time. Max delivered the top Intelligence Index score in this test, but its average cost was nearly three times High’s for three additional points, and its average task took more than twice as long. Whether that difference is worthwhile depends on the value of the task and the cost of delay.
Artificial Analysis also compared Fable 5.2 at High with Claude Opus 5.5 at Max. In its index, Fable scored 62 at $9.08 per task, while Opus scored 59 at $10.04. That finding may interest buyers weighing Anthropic’s models, but it is limited to this benchmark and its cost assumptions. Actual spending will depend on prompts, input size, output length and how often tasks are run.
Popular right now
Dreame G10 Pro Wet-Dry Vacuum Mop
Vitamix 64 oz Low-Profile Blender Container with SELF-DETECT, BPA-Free
Breville Barista Express Espresso Machine with Built-In Grinder, Black Truffle
As an affiliate, we earn on qualifying purchases.
Fable, Mythos and Model Pricing
Fable 5.2 replaces Fable 5.1, while Anthropic has also introduced Mythos 5.2 for organizations accepted into its Cyber or Life Sciences Verification Programs. The source describes Mythos as the same underlying model without some of Fable’s safety restrictions. Anthropic says Fable applies safeguards to certain high-risk cybersecurity and research-biology requests; Mythos is restricted to vetted organizations in those dual-use areas. The two names therefore refer to different access and safety arrangements, not a general price difference established by the index figures.
Anthropic says the new models produce answers up to 30% faster than their 5.1 counterparts. It also offers an optional fast mode, priced at $16 per million input tokens and $80 per million output tokens, which the company says can generate tokens up to 2.5 times faster. Those are vendor statements, separate from Artificial Analysis’s effort-level measurements. Anthropic recommends starting with Opus 5.5 for most coding and knowledge work and moving to Fable when a task calls for it.
Anthropic has also described demanding prerelease demonstrations, including a multi-agent effort to formalize a proof related to Fermat’s Last Theorem. The source identifies these as Anthropic’s own accounts, not independent verification. They do not establish how Fable will perform on ordinary business tasks. This report was drafted with help from Anthropic’s Claude, according to the source article; it states that Anthropic and Artificial Analysis did not review or sponsor it.
“The new models write answers up to 30% faster than the 5.1 versions.”
— Anthropic, describing its models
Limits of the Benchmark Figures
The cost and score figures describe Artificial Analysis’s particular index, not every prompt or production workload. The supplied material does not give a per-customer estimate, the spread of costs across individual tasks, or enough detail to predict how effort settings will affect a specific organization’s results. Scores are rounded, so High and Xhigh may differ on underlying values despite both being shown as 62.
Anthropic’s claims about faster responses, safety performance and prerelease demonstrations are company statements. The source material does not establish independent verification of those claims. Nor does it provide enough information to determine how broadly Mythos access will be granted, or how Fable’s fallback responses affect cost and performance for blocked requests.
Testing Settings on Real Workloads
Businesses considering Fable can compare effort levels on representative tasks, tracking quality, latency and token costs rather than relying on a single index score. The figures available now show the trade-off on Artificial Analysis’s test suite; they do not identify one setting as best for every use. Further benchmark updates, customer usage data and clarification from Anthropic about access and safeguards could help show whether the measured differences carry over to real deployments.
Key Questions
What does an effort level change?
It sets how much computation the model uses for a response. In Artificial Analysis’s tests, higher effort generally used more output tokens and took longer, while producing higher index scores overall.
How much does each Fable 5.2 level cost per index task?
Artificial Analysis reported averages of $3.10 at Low, $3.80 at Medium, $9.08 at High, $14.83 at Xhigh and $26.54 at Max. These are average API costs for tasks in its Intelligence Index, not fixed prices for all requests.
Is Max the best setting for every task?
No. Max scored highest on the Intelligence Index, but it took longer and cost more than the lower settings. High slightly outscored Max on the cited FrontierCode 2.0 test, illustrating that results vary by evaluation.
Are the API token prices higher at Max?
Anthropic’s listed standard pricing is the same across effort levels: $8 per million input tokens and $40 per million output tokens. The benchmark’s higher average costs at higher effort reflect greater token use, according to the supplied figures.
Who can use Mythos 5.2?
According to the source material, Mythos is available only to organizations accepted into Anthropic’s Cyber or Life Sciences Verification Programs. The material does not specify the full application criteria or how many organizations have access.
Source: ThorstenMeyerAI.com
