Key takeaways

  • Anthropic released Claude Fable 5.2 to everyone and Claude Mythos 5.2 to vetted organizations on November 9.
  • Both cost $8 per million input tokens and $40 per million output tokens, or $16 and $80 in fast mode.
  • Anthropic reports 49.1% on FrontierCode 2.0 and 74.2% on Humanity's Last Exam with tools, and still recommends Opus 5.5 for daily work.

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9. Fable 5.2 is open to anyone and costs $8 per million input tokens and $40 per million output tokens, which is 20% less than Fable 5.1, according to The New Stack. Mythos 5.2 is the same underlying model without the cyber and biology safeguards, and only vetted organizations can use it.

Anthropic calls the pair "best for the most ambitious, open-ended technical work." It also says that for most coding and knowledge work, Claude Opus 5.5 is still the model to use every day. That is an unusual thing for a lab to say about its own flagship, and it tells you who the new models are for: people with hard problems, and budgets to match.

What's new in Fable 5.2 and Mythos 5.2

Fable 5.2 is available today on all platforms, including Amazon Web Services, Google Cloud and Microsoft Azure. The API model ID is claude-fable-5-2. Anthropic says outputs run up to 30% faster than Fable 5.1 and Mythos 5.1. AWS has already announced both models on Bedrock.

Fast mode generates tokens up to 2.5 times faster. It runs in Claude Code, the Claude Platform, Claude Models on Google Cloud and Microsoft Foundry. Mythos fast mode is limited to verified organizations.

Anthropic also says the models formalized a proof of Fermat's Last Theorem and solved two open black hole physics problems during prerelease testing. Those are Anthropic's claims, and the company has not published outside verification in the announcement.

Who can use Mythos 5.2

Mythos 5.2 has no cyber or biology safeguards, so access is gated:

  • Cybersecurity professionals can apply to the Cyber Verification Program.
  • Life sciences researchers can apply to the Life Sciences Verification Program.
  • Claude Enterprise customers can use Claude Security, which runs on Mythos 5.2, without verification.

Fable 5.2 blocks about 55% fewer requests than Fable 5.1, Anthropic says: about 60% fewer in cyber and 40% fewer in biology. A blocked cyber request falls back to Opus 4.8, and a blocked biology request falls back to Opus 5.

Pricing

Input, per million tokens Output, per million tokens
Fable 5.2 and Mythos 5.2 $8 $40
Fast mode $16 $80

The announcement does not state a context window. OpenRouter lists Fable 5.2 with 1,000,000 tokens of context at the same $8 and $40 prices, and a batch version at $4 and $20.

Eligible paid plans get one to three Fable resets, depending on how long the subscription has run. Fable covers up to 50% of weekly usage limits for Pro and standard Team seats. Anthropic's launch promotion article has the details.

Benchmarks, as Anthropic reports them

These are Anthropic's own numbers for Fable 5.2 unless noted. Independent results are still coming in.

Benchmark Fable 5.2 Comparison
FrontierCode 2.0, Xhigh effort 49.1% Highest of any released model Anthropic tested
FrontierCode 2.0, High effort about 45.3% Opus 5.5: 45.5%. Anthropic cites cost per task about half
Humanity's Last Exam, with tools 74.2% Opus 5.5: 67.7%, GPT-6 Astra: 57.2%
OSWorld 2.1, computer use 89.3% Partial offline set
Remote Labor Index, High effort 41.3% Opus 5.5: 21.3%
Chartography, no tools 73.4% Gemini 4 Carbon: 72.5%

For Mythos 5.2, Anthropic reports 89.6% on CVE Rediscovery, a test of finding known vulnerabilities. It says that is 10.2% more vulnerabilities found than Opus 5.5.

Artificial Analysis has listed Fable 5.2 at 65 on its index, the highest so far on that leaderboard. It was one of the top posts on Hacker News on Monday, with more than 700 points.

Safety notes

Anthropic rates the risk of high-stakes misaligned behavior as low. In a sandbox-escape test, Mythos 5.2 used out-of-sandbox credentials or files in 3.1% of scenarios, against 21% for Mythos 5.1. Both models sometimes suspect they are being tested. The UK AI Security Institute, Frontier Design and METR were external testers, and Anthropic notes they did not review the blog post or the system card.

What to watch

Third-party benchmarks and hands-on reports over the next few days will show whether the gap over Opus 5.5 holds outside Anthropic's own tests. Pricing is the other thing to watch: at High effort Fable 5.2 roughly ties Opus 5.5 on FrontierCode, so the case for it rests on the hardest tasks and on cost per task. For prices on the older models, see our Claude Opus 5.5 pricing page.

Sources

  1. Claude Fable 5.2 and Mythos 5.2Anthropic
  2. Claude Fable 5.2 and Mythos 5.2 system cardAnthropic
  3. Introducing Claude Fable 5.2 on AWSAWS
  4. Anthropic: Claude Fable 5.2OpenRouter
  5. Claude Fable 5.2 on Artificial AnalysisArtificial Analysis

Add ClaudeGPT Si as a preferred source on Google