🔍 Read the full analysis: The Cost Of Thinking Harder: Claude Fable 5.2 per-Task Pricing on ThorstenMeyerAI.com
Get audio and creator gear 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 trade task cost and speed for higher scores on Artificial Analysis’s Intelligence Index. The independent test reports average costs from $3.10 at Low to $26.54 at Max; Anthropic also released a separate, access-restricted model called Mythos 5.2. The benchmark does not establish how the cost and performance trade-offs will play out on every business’s workload.
Anthropic’s Claude Fable 5.2 can score from 54 to 65 on Artificial Analysis’s Intelligence Index depending on its effort setting, while the reported average cost per test task ranges from $3.10 to $26.54. The results, accessed November 10, put Fable at the top of the 74-model ranking at Max effort, but show businesses face a substantial cost and time increase for the highest score.
Artificial Analysis tested five settings—Low, Medium, High, Xhigh and Max—on its Intelligence Index, which covers nine evaluations in areas including agentic coding, business documents, science and knowledge. At Low, Fable scored 54, ranked 15th, and averaged $3.10 and 6.5 minutes per task. The default Medium setting scored 56, ranked ninth, and averaged $3.80 and 8.3 minutes.
At High, the model scored 62 for $9.08 per task, taking an average of 17.9 minutes. Xhigh also scored 62 as shown by the benchmark, at $14.83 and 26.4 minutes. Max reached 65 and first place, with an average task cost of $26.54 and a 46.4-minute runtime. These are benchmark averages, not a guarantee of what a particular customer’s request will cost or take.
Anthropic lists the same standard token prices for all five effort levels: $8 per million input tokens and $40 per million output tokens. The cost difference in the test comes from how much output the model uses while working: Artificial Analysis reports 34.9 million output tokens across its index at Low and 275 million at Max. Anthropic says Fable 5.2 writes answers up to 30% faster than version 5.1. Its optional fast mode is priced at $16 per million input tokens and $80 per million output tokens, with Anthropic saying it can generate tokens up to 2.5 times faster.
The Cost of Thinking Harder Claude Fable 5.2 per-task pricing
Higher effort settings scored better on Artificial Analysis’s Intelligence Index, but also took longer and cost more per test task. These benchmark averages may not predict costs on your business’s own workload.
Cost and score by effort
Index results show the trade-off across five settings; costs and times are benchmark averages per task.
| Effort | Score | Cost | Time |
|---|---|---|---|
| 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 higher effort pays—and may not
More effort can raise the index score, but added cost and time do not guarantee better results on every evaluation.
- Medium to High adds six score points for $5.28 more per benchmark task.
- Max reaches the top index score, with the highest reported cost and runtime.
- High and Xhigh both display a score of 62, but Xhigh costs more.
- On FrontierCode 2.0, High scored 49.1% and Max 48.8%.
Choose the model for the task
Anthropic positions Fable for work that warrants it, while access and safeguards differ between the two models.
Claude Fable 5.2
Available generally, with safeguards for some high-risk cybersecurity requests and most research biology tasks.
Mythos 5.2
Described as the same underlying model without those safeguards; access is limited to accepted verification-program organizations.
Opus 5.5
Anthropic recommends it for most coding and knowledge work, with Fable for tasks that need it.
Test against your own workload
The index offers a starting point, not a forecast of your company’s spend or business value.
Use the real prompts and work your team needs to complete.
Track quality, API cost, and latency together.
Use Xhigh or Max when their results justify the extra cost and wait.
The Price of Fable’s Highest Score
The benchmark makes the effort setting a practical purchasing decision, not just a quality preference. Moving from Medium to High costs $5.28 more per task in Artificial Analysis’s test and adds six points to the rounded score. Moving from High to Max costs nearly three times as much for three additional points, while the reported average runtime rises from 17.9 to 46.4 minutes.
That may matter for teams running large numbers of requests or waiting on interactive coding and research tasks. Xhigh’s displayed score is the same as High’s, despite a 63% higher benchmark task cost. The results suggest that using the most expensive setting for every request may not be necessary, but a benchmark cannot determine the right setting for a company’s own work.
Artificial Analysis also reported a result that cautions against treating more effort as an automatic improvement: on its FrontierCode 2.0 agentic coding test, Fable scored 49.1% at High and 48.8% at Max. That is a result from a specific evaluation, not evidence that High will outperform Max across tasks.
Products Worth Considering
LG UltraGear 27-inch QHD OLED Gaming Monitor, 240Hz
LG UltraGear 25G590B-B 25-Inch Full HD 1000Hz Gaming Monitor
Bose QuietComfort Headphones (2nd Gen), Wireless Over-Ear Noise-Cancelling Headphones, Black
As an affiliate, we earn on qualifying purchases.
Two Models, Different Access
Anthropic announced Fable 5.2 and Mythos 5.2 on November 9, replacing their respective 5.1 versions. According to the source material, Fable is available to users generally, with safeguards that block some high-risk cybersecurity requests and most tasks in research biology. Queries blocked by those safeguards receive responses from older models, identified as Opus 4.8 for cybersecurity and Opus 5 for biology.
Anthropic describes Mythos 5.2 as the same underlying model with those safeguards removed, but restricts access to organizations accepted into its Cyber or Life Sciences Verification Programs. Anthropic says early testing showed about 55% fewer blocks with Fable 5.2 than with Fable 5.1. That is a company-reported result, and the source material does not provide independent verification.
Anthropic also recommends that most customers start with Claude Opus 5.5 for coding and knowledge work, moving to Fable when a task warrants it. In Artificial Analysis’s comparison, Opus 5.5 at Max scored 59 at $10.04 per task, while Fable at High scored 62 at $9.08. Those figures come from the index and should not be treated as a universal comparison across workloads.
What Benchmark Costs Cannot Tell You
Artificial Analysis’s per-task figures are average API costs for tasks in its nine-evaluation Intelligence Index. They do not establish what an individual company will spend on its own prompts, how often it will need each effort level, or whether the score differences will translate into business value. The source material does not provide a measured return on investment.
Some of the most striking capability examples in the source—including work on electronic devices, black-hole physics and a proof effort involving Fermat’s Last Theorem—are Anthropic’s accounts of prerelease testing, not independently verified outcomes. Anthropic’s comparison of that proof effort with Opus 5.5 was based on one run per model with about 1,000 agents each. The source also notes that both teams resumed from their last save at the 18-hour point.
The article supplying the figures states that it was drafted with help from Claude, that Anthropic is the subject, and that neither Anthropic nor Artificial Analysis reviewed or sponsored it. The underlying benchmark results and Anthropic’s claims should be read with those disclosures and their different levels of verification in mind.
Testing Fable Against Real Workloads
The next practical step for businesses is to compare effort settings on their own tasks, tracking quality, latency and API cost together rather than selecting a setting from the headline ranking alone. The reported results support trying High where a task needs more capability than Medium provides, while checking whether Xhigh or Max produces enough additional value to justify its higher cost and longer wait.
Anthropic’s guidance points customers toward Opus 5.5 for much routine coding and knowledge work, with Fable reserved for tasks that require it. The source material does not give a date for further releases, a broader independent evaluation of Mythos, or additional verification of Anthropic’s prerelease demonstrations. Those details remain open.
Key Questions
What is the reported cost of Claude Fable 5.2 per task?
In Artificial Analysis’s Intelligence Index, average API cost ranges from $3.10 at Low to $26.54 at Max. These are benchmark averages, not a fixed price for every request.
Which effort setting ranked highest?
Max effort scored 65 and ranked first among 74 models on the index as reported on November 10, 2026. High and Xhigh each displayed a score of 62.
Why do Fable’s effort settings have different costs?
Anthropic lists the same standard input and output token prices across the settings. In the benchmark, higher effort used more output tokens, raising the average task cost and runtime.
How does Fable 5.2 differ from Mythos 5.2?
According to the source material, Fable includes safeguards for some cybersecurity and biology requests and is available to users generally. Mythos is described as the same underlying model without those blocks and is limited to organizations accepted into Anthropic’s verification programs.
Does the benchmark show that Max is best for every task?
No. It shows Max had the highest Intelligence Index score, but not that it is best for every workload. On the reported FrontierCode 2.0 test, High scored 49.1% and Max scored 48.8%.
Source: ThorstenMeyerAI.com
