# Claude Fable 5.2 scores 65 on Artificial Analysis at $8/$40 per million tokens

> Anthropic's Fable and Mythos 5.2 rank first on the Artificial Analysis Intelligence Index, and Anthropic's own table puts them ahead of Opus 5.5 on most benchmarks. At Max effort the model is slow and verbose, so the effort setting decides the bill.

- Published: 2026-11-09
- By: Samir Sengupta (https://www.samcodeman.com/about)
- Canonical: https://www.samcodeman.com/writing/claude-fable-5-2-65-artificial-analysis-8-40-pricing
- How this was made: Drafted with Claude, Anthropic's model from 2 cited sources, then checked against them in a second editing pass. Editorial policy: https://www.samcodeman.com/editorial-policy
- Disclosure: I build Wayne, an AI coding harness for VS Code, and Omnitron, an agent harness, so I work in this category.

## The short version

- Fable and Mythos are the same underlying model. Fable carries cybersecurity and biology safeguards and is open to all users, while Mythos requires an application through Anthropic's verification programs.
- In Anthropic's charts, Fable 5.2 at lower effort settings matches Opus 5.5's best scores on several benchmarks. The cost per task ranges from roughly half to less than one-tenth of Opus 5.5's.
- At Max effort, Artificial Analysis measured 46.4 minutes and $26.54 per Intelligence Index task, which fits unattended jobs better than interactive loops.
- Anthropic still recommends Opus 5.5 as the starting model and reserves Fable or Mythos for the most ambitious, open-ended technical work.

Anthropic released Claude Fable 5.2 and its gated counterpart, Claude Mythos 5.2, on November 9, 2026. Both cost $8 per million input tokens and $40 per million output tokens. Artificial Analysis scores the Fable max configuration at 65 on its Intelligence Index, first of 74 models, and lists a 1M-token context window. Anthropic's own benchmark table puts the new models ahead of Opus 5.5 on almost all the capabilities it evaluated. Anthropic still tells most users to start with Opus 5.5 and move up only for the hardest work.

Anthropic says the new models generate output up to 30% faster than Fable 5.1 and Mythos 5.1, and that they are priced lower than that generation. An optional fast mode speeds token generation by up to 2.5 times. Artificial Analysis lists text and image input and text output for Fable. Anthropic also positions Fable in two supporting roles: as an Advisor that Haiku, Sonnet or Opus models consult on harder problems, and as an Orchestrator that delegates simpler tasks to smaller models.

## Fable and Mythos are one model, gated differently

Anthropic states that Fable and Mythos are the same underlying model. Fable carries strong cybersecurity and biology safeguards, which is why Anthropic can make it available to all users. Organizations pursuing beneficial applications in cybersecurity or biomedical research can apply for Mythos through the Life Sciences and Cyber Verification Programs. Anthropic notes that Fable's safeguards will block most cybersecurity-related tasks. It says those safeguards are as robust as Fable 5.1's but produce fewer unnecessary blocks.

Anthropic reports no Fable 5.2 scores on its security rows. The comparisons it gives are between Mythos 5.2, Mythos 5.1 and Opus 5.5:

- On CVE Rediscovery, Mythos 5.2 scores 89.6%, against 81.3% for Opus 5.5 without safeguards and 78.3% for Mythos 5.1. Anthropic says that works out to 10.2% more vulnerabilities found than Opus 5.5 and 14.4% more than Mythos 5.1.
- On CyScenarioBench, Mythos 5.2 scores 76.3%, against 67.6% for Opus 5.5 without safeguards and 61% for GPT-6 Cyber. Anthropic notes that each lab's scores on this benchmark rest on unique tasks.
- One security company found that one in six confirmed bugs found by Mythos 5.2 were missed by all the other frontier models it tested.

The only row that reports both tiers is Terminal-Bench-Science 0.1, where Fable scores 74.6% and Mythos 75.3%. Anthropic positions the Mythos application process around cybersecurity and biomedical work. Teams whose pipelines touch vulnerability research should plan for Fable refusing most of that work.

## Price is $8/$40, but effort sets the bill

Standard pricing is $8 per million input tokens and $40 per million output tokens. Fast mode costs $16 and $80. Artificial Analysis classes the standard rate as expensive against medians of $2.00 and $10.00, and lists a 98% cache discount. That discount matters for long, repeated prompts. Artificial Analysis recorded 28.3B input tokens for the model across its index.

Verbosity is the bigger cost driver. At Max effort, the model generated 275M tokens across the Intelligence Index, against a median of 72.8M. That came to $26.54 per task, which ranks 73rd of 73 on cost. The sources do not give per-token prices for Fable 5.1 or Opus 5.5, so a per-token comparison with either is not possible from this material. The only cost comparisons available are Anthropic's per-task charts.

## When Claude Fable 5.2 beats Opus 5.5 on cost

Anthropic's per-task charts show the new Fable matching or passing Opus 5.5's best results at lower effort settings:

- FrontierCode 2.0: at High effort it scores 45.3% against Opus 5.5's best of 45.5%, for roughly half the cost per task. It also beats GPT-6 Astra's best of 42.8% at about the same cost. At Xhigh effort it reaches 49.1%, the highest of any released model Anthropic tested.
- WANDR: at Medium effort it matches Opus 5.5's top score for about 40% of the cost per task.
- Humanity's Last Exam (with tools): at Low effort it matches Opus 5.5's top score for less than one-tenth of the cost per task.
- Chartography: at Medium effort it scores 66.4%, ahead of Opus 5.5 at Max effort (64.4%), for less than one-tenth of the cost per task.

Medium is the default effort in Claude apps and Claude Code. Anthropic says Fable 5.2 completes challenging tasks there with fewer tokens than Fable 5.1. The Artificial Analysis cost and speed figures describe the Max configuration only. Teams should measure the effort levels they actually plan to run rather than rely on the headline index.

## Max effort takes 46.4 minutes per task

Artificial Analysis ranks the Max configuration 71st of 72 on performance. It measured 215.95 seconds to the first answer token, an output speed of 98 tokens per second, and 46.4 minutes per Intelligence Index task, against a median of 12.7 minutes. That profile is a poor fit for interactive loops and suits long unattended jobs.

Anthropic's own examples are unattended jobs. In one, the model coordinated more than 1,000 subagents over eight days to build a prototype operating system that boots on a laptop. In another, a team of agents running an earlier version of Fable 5.2 migrated a production service from Python to Rust, and the service now starts about 26 times faster. An Uber staff engineer described a 10 to 12 hour security audit in which the model sent out about 170 reviewer agents. He said he would use it at High effort for long unattended audits:

> It works until the job is actually complete instead of stopping at a plausible checkpoint.
>
> - Andrew Mager, Uber, quoted by Anthropic

## Where Fable 5.2's lead is thin or reversed

Anthropic's table supports its overall claim, but the margins vary widely:

- Large gains over Opus 5.5: Remote Labor Index 41.3% vs 21.3%, Terminal-Bench-Science 0.1 74.6% vs 52.6%, RoboDojo Sim 64.9% vs 39.4% (run internally), BenchCAD 2.1 83.5 vs 65.3 at Max effort, and FrontierSWE v2 69.9% vs 62.3%.
- Small gains over Opus 5.5: OSWorld 2.1 89.3% vs 87.2%.
- Not first: on AutomationBench, Gemini 4 Carbon leads at 53.9% against Fable's 51.4%, with Opus 5.5 at 42.5%. On AA-Briefcase v1.1, Opus 5.5 has the higher Elo, 1800 against 1785, although Fable leads on rubric score, 66.1% to 61.2%.

Commenters on Hacker News asked whether the Opus 5.5 numbers come from launch or reflect the model's current behavior, which some believe has degraded. The sources do not say which Opus snapshot or test date Anthropic used. Anthropic states only that scores show each model's highest-available effort, and that Fable 5.2, Fable 5.1 and Opus 5.5 run with safeguards on unless noted.

One commenter objected that FrontierCode 2.0 forces Gemini to use Devin CLI while other models use their preferred harness. Anthropic's post does not describe per-model harnesses, so this note cannot confirm or refute that objection.

That note looked at how models tied on a coding benchmark split apart under task mutation. It is a reason to treat narrow table margins, such as the OSWorld gap, with caution. See: [MaxxBench benchmark splits DeepSWE's 74% cluster into 53.4% to 76.4%](https://www.samcodeman.com/writing/maxxbench-benchmark-splits-deepswe-74-percent-cluster)

## When to pick Fable or Mythos over Opus 5.5

1. Start with Opus 5.5 for most coding and knowledge tasks, as Anthropic recommends. Move down to Sonnet 5.5 or Haiku 5.5 where the scope allows.
2. Move up to Fable for open-ended, long-horizon work like the examples Anthropic cites: software migrations, multi-agent orchestration and hardware design. Test it at Low to High effort before paying for Max.
3. Use Fable as an Advisor to Haiku, Sonnet or Opus when only a few steps need its reasoning, rather than paying its output rates for the entire run.
4. Apply for Mythos only if the work involves cybersecurity or biomedical research, where Fable's safeguards will block most tasks.

Several questions remain open. Commenters were puzzled that the flagship carries a lower version number than Opus 5.5. One commenter argued that the different pricing and speed suggest a different architecture from Fable 5.1. Another, citing the system card, said this release is METR's Model 3 and that a stronger model is waiting. Anthropic's post does not address the naming or any next model.

On writing quality, Anthropic says testers found the new models communicate more clearly than earlier Fable and Mythos releases. Commenters disagreed about Claude's prose in general, and one said they had not yet tried this version.

## Questions this raises

### What is the difference between Claude Fable 5.2 and Claude Mythos 5.2?

Anthropic says they are the same underlying model. Fable carries strong cybersecurity and biology safeguards and is available to all users, while Mythos is gated through the Life Sciences and Cyber Verification Programs. Fable's safeguards will block most cybersecurity-related tasks.

### How much does Claude Fable 5.2 cost to run?

Standard pricing is $8 per million input tokens and $40 per million output tokens, with a 98% cache discount listed by Artificial Analysis. At Max effort the model is verbose, averaging $26.54 per Intelligence Index task. Anthropic's charts show lower effort settings matching Opus 5.5's best results at a fraction of the cost per task.

### Does Claude Fable 5.2 beat Opus 5.5 on every benchmark?

No. Gemini 4 Carbon leads AutomationBench at 53.9% against Fable's 51.4%, and Opus 5.5 has a higher Elo on AA-Briefcase v1.1, 1800 vs 1785. Some gains are narrow, such as OSWorld 2.1 at 89.3% vs 87.2%.

## Key facts

- Claude Fable 5.2 and Mythos 5.2 launched November 9, 2026 at $8 input and $40 output per million tokens; fast mode costs $16/$80.
- Artificial Analysis scores Fable 5.2 max at 65 on its Intelligence Index, first of 74 models, with a 1M-token context window.
- At Max effort it generated 275M tokens across the index (median 72.8M), costing $26.54 per task and ranking 73rd of 73 on cost.
- Max effort takes 46.4 minutes per index task (median 12.7) and 215.95 seconds to first answer token.
- At High effort on FrontierCode 2.0 it scores 45.3% vs Opus 5.5's best of 45.5% for roughly half the cost per task.

## Sources

1. [Claude Fable 5.2 and Mythos 5.2](https://www.anthropic.com/claude-fable-and-mythos-5-2) - anthropic.com
2. [Claude Fable 5.2 Scores 65 on Artificial Analysis, Highest So Far](https://artificialanalysis.ai/models/claude-fable-5-2) - artificialanalysis.ai

Discussion read (practitioners' reactions, not a source): https://news.ycombinator.com/item?id=50429906