🔍 Read the full analysis: Claude Fable 5.2 Takes The #1 Spot: What Effort Costs You Per Task on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Anthropic released Claude Fable 5.2 on Nov. 9, 2026, and Artificial Analysis ranked its Max effort setting first among 74 models the next day. The ranking comes with a wide price range: the benchmark’s estimated cost per task runs from $3.10 at Low to $26.54 at Max, while High scores 62 for $9.08. These are benchmark results, not a guarantee of performance or cost on a business’s own workload.
Anthropic released Claude Fable 5.2 on Nov. 9, and the model’s Max effort setting ranked No. 1 of 74 on the Artificial Analysis Intelligence Index a day later. The benchmark estimates a cost of $26.54 per task at Max, compared with $9.08 at High and $3.10 at Low, giving businesses a price-performance trade-off behind the top ranking.
Artificial Analysis reported a score of 65 at Max effort. Fable 5.2 also took the next two positions on the ranking: Xhigh and High each scored 62, at estimated costs of $14.83 and $9.08 per task, respectively. Medium, the default in Claude apps and Claude Code, scored 56 at $3.80; Low scored 54 at $3.10. The index covers nine evaluations involving agentic coding, business documents, science and knowledge work.
The reported task cost is the average API cost for a task in that evaluation, not a universal price for every request. Artificial Analysis listed the same token rates for each effort level: $8 per million input tokens and $40 per million output tokens. The difference in estimated bills 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 higher setting also took longer in the benchmark: 46.4 minutes per task at Max, compared with 17.9 minutes at High and 6.5 minutes at Low. The move from Medium to High added six index points for $5.28 more per task. Xhigh cost 63% more than High while showing the same rounded score; Max cost nearly three times as much as High for three additional points. These figures describe the index’s reported results, not an expected outcome for every customer.
Claude Fable 5.2 takes the lead What effort costs per task
Artificial Analysis ranked Max effort first among 74 models. The top score comes with higher estimated cost and longer task time.
Effort changes the trade-off
Index scores and average task costs vary by setting.
| Effort | Score | Cost per task |
|---|---|---|
| Max | 65 | $26.54 |
| Xhigh | 62 | $14.83 |
| High | 62 | $9.08 |
| Medium | 56 | $3.80 |
| Low | 54 | $3.10 |
What the extra effort uses
The token rates are the same across effort levels; estimated bills differ with token use and time.
- 34.9 million output tokens across the index
- 6.5 minutes per task
- 275 million output tokens across the index
- 46.4 minutes per task
Ranking has limits
Results come from nine evaluations spanning agentic coding, business documents, science and knowledge work.
Not a universal bill
Task costs are averages from the evaluation, not guaranteed prices for a business request.
Max did not lead
In FrontierCode 2.0, High scored 49.1% and Max 48.8%.
Choose for the workload
Compare quality, cost, token use and completion time on the tasks your team actually runs.
Check whether it meets the task’s quality and response-time needs.
Measure the added score against the added cost and time.
Anthropic recommends starting with Opus 5.5 for most coding and knowledge work.
The Price of the Top Ranking
The results make effort level a budget and latency choice, not just a model setting. A team running many tasks could see a substantial difference between the benchmark’s $9.08 High estimate and $26.54 Max estimate. The additional time at Max may also matter when people are waiting for a response or when a workflow has a deadline.
The index does not show that Max is the right setting for every difficult job. In Artificial Analysis’s separate FrontierCode 2.0 coding test, Fable 5.2 scored 49.1% at High and 48.8% at Max. The result is a reminder that more inference effort did not improve performance in that particular test. Businesses would need to evaluate their own workloads, quality requirements and response-time limits before choosing a setting.
The source also compares Fable 5.2 at High with Anthropic’s Claude Opus 5.5 at Max: Artificial Analysis lists scores of 62 for $9.08 versus 59 for $10.04. That comparison may help frame model selection, but it does not establish how either system will perform across a company’s specific tasks. The benchmark’s cost and scores should be treated as historical test results, not price or performance guarantees.
Products Worth Considering
Apple MacBook Air 13.6-inch Laptop with M5 Chip, 16GB Memory, 512GB SSD, Midnight
Apple MacBook Neo 13-inch Laptop with A18 Pro Chip, 8GB Memory, 256GB SSD, Liquid Retina Display, and 1080p Camera — Blush
Logitech MX Mechanical Wireless Illuminated Keyboard, Clicky Switches
As an affiliate, we earn on qualifying purchases.
Fable and Mythos Serve Different Users
Anthropic introduced Fable 5.2 and Mythos 5.2 as replacements for the respective 5.1 models. According to the company, the two new names refer to the same underlying model, with different access and safety controls. Fable is available to users with safeguards for high-risk cybersecurity and many research-biology requests. Mythos lacks those blocks and is restricted to organizations accepted into Anthropic’s Cyber or Life Sciences Verification Programs.
Anthropic says Fable answers up to 30% faster than the 5.1 versions. Its optional fast mode generates tokens at up to 2.5 times the speed, with listed rates of $16 per million input tokens and $80 per million output tokens. Anthropic also says early testing showed about 55% fewer blocked requests than with Fable 5.1. These are company claims; the supplied material does not include independent verification of them.
Anthropic described prerelease demonstrations including electronic-device designs, black-hole physics problems and a multi-agent attempt to formalize Fermat’s Last Theorem. The company reported that, in one run using about 1,000 agents, Fable 5.2 proved all weighted statements in 30 hours, while Opus 5.5 reached 27% after 30 hours. Those results are Anthropic’s own account, based on a single run per model, and should not be read as independent evidence of typical customer performance.
“The company recommends starting with Claude Opus 5.5 for most coding and knowledge work, and moving to Fable when a task needs it.”
— Anthropic
Benchmark Limits and Open Questions
The index figures are benchmark estimates, not observed bills for all users. The source material does not specify how representative the nine evaluations are of common commercial workloads, or how much task costs vary across different prompts and deployment setups. It also does not provide enough detail to determine whether a business would see the same score differences at each effort level.
The supplied information does not include independent confirmation of Anthropic’s speed, safety-blocking or prerelease-demonstration claims. The reported Fermat result rests on one run per model, and the source does not establish that it can be reproduced. It is also not clear how the Mythos access program will operate in practice for applicants, beyond Anthropic’s requirement that organizations be vetted.
Testing Costs on Real Workloads
The next useful test for prospective customers is whether the benchmark trade-off holds on their own tasks. Teams can compare quality, token use, cost and completion time across effort levels, while checking whether High handles the work before paying for Max. Anthropic’s recommendation to begin with Opus 5.5 for many coding and knowledge tasks offers another comparison point, though its fit will depend on the job.
Further detail on independent testing, repeatable results and access to Mythos could clarify how much of the reported performance carries over beyond the index. Until then, the No. 1 ranking is confirmed only for the cited benchmark; the best setting for a particular organization remains an open, workload-specific question.
Key Questions
What does Claude Fable 5.2’s No. 1 ranking mean?
Artificial Analysis ranked Fable 5.2 at Max effort first among 74 models on its Intelligence Index, with a score of 65. The rank applies to that benchmark and does not establish that it is best for every use case.
How much does a Fable 5.2 task cost?
Artificial Analysis estimated an average index-task cost of $3.10 at Low, $9.08 at High and $26.54 at Max. Actual costs depend on usage; these figures are benchmark estimates, not a guaranteed charge for a particular request.
Is Max effort always better than High?
No. Max scored 65 versus 62 for High on the Intelligence Index, but it cost nearly three times as much per task in that test. On the separate FrontierCode 2.0 coding test, High scored 49.1% and Max 48.8%.
What is the difference between Fable 5.2 and Mythos 5.2?
Anthropic says they share the same underlying model, but Mythos is restricted to vetted organizations in its Cyber or Life Sciences Verification Programs and does not have the same safety blocks. Fable is the broadly available version with safeguards for certain high-risk requests.
Are Anthropic’s prerelease achievement claims independently verified?
The supplied source describes the device-design, physics and Fermat’s Last Theorem results as Anthropic’s own accounts. It does not provide independent verification, and the reported Fermat comparison involved one run per model.
Source: ThorstenMeyerAI.com
