Hermes Vector

AI & SI

Anthropic Introduces Claude Fable 5.2 and Claude Mythos 5.2

Anthropic has released upgraded versions of Claude Fable and Claude Mythos, models designed for technical challenges including coding, reasoning, and vision.

By Hermes Vector

Checked against 1 sourceAnthropic
Illustration for: Anthropic Introduces Claude Fable 5.2 and Claude Mythos 5.2

In brief

  • Anthropic introduced Claude Fable 5.2 and Claude Mythos 5.2, upgraded models for technical challenges.
  • Fable 5.2 and Mythos 5.2 generate outputs up to 30% faster than 5.1 versions at $8 per million input tokens and $40 per million output tokens.
  • Fable 5.2 is available to all users, while Mythos 5.2 access requires application through Life Sciences and Cyber Verification Programs.
  • Anthropic reports the models formalized Fermat’s Last Theorem in 30 hours and solved two open problems in black hole physics.

Anthropic Introduces Claude Fable 5.2 and Claude Mythos 5.2

Anthropic has introduced Claude Fable 5.2 and Claude Mythos 5.2, upgraded models designed to tackle technical challenges. The company states that Claude Fable and Claude Mythos are the same underlying model, with Fable featuring cybersecurity and biology safeguards that allow it to be available to all users. Organizations can apply for access to Mythos through the Life Sciences and Cyber Verification Programs.

Performance and Pricing

Anthropic says testers found that Fable 5.2 and Mythos 5.2 communicate more clearly than past models. They generate outputs up to 30% faster than Claude Fable 5.1 and Claude Mythos 5.1, and are available at $8 per million input tokens and $40 per million output tokens. An optional fast mode speeds up token generation by up to 2.5 times and is available for $16 per million input tokens and $80 per million output tokens.

Benchmark Results

According to Anthropic, the benchmarks show Claude Fable 5.2 and Mythos 5.2 are a step above Opus 5.5 in almost all capabilities evaluated. The following scores are from the source:

  • Agentic coding (FrontierCode 2.0): Fable 5.2 at 49.1%, Fable 5.1 at 32.4%, Opus 5.5 at 45.5%, GPT-6 Astra at 42.8%, Gemini 4 Carbon at 41.1%
  • Agentic coding (FrontierSWE v2): Fable 5.2 at 69.9%, Fable 5.1 at 56.3%, Opus 5.5 at 62.3%, GPT-6 Astra at 65.5%, Gemini 4 Carbon at 61.2%
  • Business workflows (AutomationBench): Fable 5.2 at 51.4%, Fable 5.1 at 31.4%, Opus 5.5 at 42.5%, GPT-6 Astra at 41.4%, Gemini 4 Carbon at 53.9%
  • Agentic scientific research (Terminal-Bench-Science 0.1): Fable 5.2 at 74.6%, Mythos 5.2 at 75.3%, Opus 5.5 at 52.6%, GPT-6 Astra at 58.7%, Gemini 4 Carbon at 68.1%
  • Cybersecurity (CyScenarioBench): Fable 5.2 at 76.3%, Mythos 5.2 at 61.7%, Mythos 5.1 at 61.7%, Opus 5.5 at 67.6% without safeguards, GPT-6 Cyber at 61%
  • Cybersecurity (CVE Rediscovery): Mythos 5.2 at 89.6%, Mythos 5.1 at 78.3%, Opus 5.5 at 81.3% without safeguards
  • Protein design (De novo binder design 24 hr): Mythos 5.2 at 85.1, Mythos 5.1 at 79.1, Opus 5.5 at 82.6 without safeguards
  • Agentic search (WANDR relaxed score): Fable 5.2 at 82.1%, Fable 5.1 at 68.7%, Opus 5.5 at 72.3%
  • Computer use (OSWorld 2.1 offline set partial): Fable 5.2 at 89.3%, Fable 5.1 at 83.5%, Opus 5.5 at 87.2%, GPT-6 Astra at 79.5%, Gemini 4 Carbon at 81.5%
  • Robotics (RoboDojo Sim): Fable 5.2 at 64.9%, Fable 5.1 at 27.6%, Opus 5.5 at 39.4%, GPT-6 Astra at 29.8%
  • Knowledge work (AA-Briefcase v1.1 rubric): Fable 5.2 at 66.1%, Fable 5.1 at 61.5%, Opus 5.5 at 61.2%, GPT-6 Astra at 52.0%, Gemini 4 Carbon at 61.8%
  • Knowledge work (1785 Elo): Fable 5.2 at 1669 Elo, Fable 5.1 at 1800 Elo, Opus 5.5 at 1569 Elo, GPT-6 Astra at 1611 Elo
  • Knowledge work (Remote Labor Index): Fable 5.2 at 41.3% High effort, Fable 5.1 at 17.9%, Opus 5.5 at 21.3%, GPT-6 Astra at 20.8%
  • Multidisciplinary reasoning (Humanity’s Last Exam with tools): Fable 5.2 at 74.2%, Fable 5.1 at 65.6%, Opus 5.5 at 67.7%, GPT-6 Astra at 57.2%
  • Visual chart recognition (Chartography no tools): Fable 5.2 at 73.4%, Fable 5.1 at 44.8%, Opus 5.5 at 64.4%, GPT-6 Astra at 71.0%, Gemini 4 Carbon at 72.5%

Applications and Safety

Anthropic reports that during prerelease testing, the models helped engineers complete software migrations, designed three working electronic devices from scratch, solved two open problems in the physics of black holes, made substantial progress on robotics evaluations, and formalized the proof of Fermat’s Last Theorem in 30 hours.

Regarding safety, Anthropic states it ran extensive prerelease safety testing, including novel evaluations for behaviors like sandbox escapes. The company believes high-stakes risks from misalignment remain low, and the model performed as well as, or better than, recent models on most alignment measures. Fable 5.2’s cyber and biology safeguards are described as robust as those of Fable 5.1 but producing fewer unnecessary blocks, with strongest-ever protections against distillation attacks implemented.

Why it matters

The release provides updated performance and pricing for models positioned for ambitious technical work, with specific access programs for cybersecurity and biomedical research applications.


Sources

This article was drafted with AI assistance and checked against the sources above. Company claims are reported as claims. Cover image is AI-generated.

Questions readers ask

What are the new models introduced by Anthropic?
Anthropic has introduced Claude Fable 5.2 and Claude Mythos 5.2, which are upgraded models designed to tackle technical challenges.
How much do Claude Fable 5.2 and Claude Mythos 5.2 cost?
The models are available at $8 per million input tokens and $40 per million output tokens. An optional fast mode is available for $16 per million input tokens and $80 per million output tokens.
How do the new models compare to previous versions in terms of speed?
Testers found that Fable 5.2 and Mythos 5.2 generate outputs up to 30% faster than Claude Fable 5.1 and Claude Mythos 5.1.
What benchmarks did Claude Fable 5.2 achieve in cybersecurity?
In the Cybersecurity (CyScenarioBench) benchmark, Fable 5.2 scored 76.3%, and in the Cybersecurity (CVE Rediscovery) benchmark, Mythos 5.2 scored 89.6%.

The day's checked news, by email

One short email a day with the top stories. Free, and you can unsubscribe in one click.

Or follow for the latest news:

How do you feel about this story?

Comments 0

No comments yet. Start the discussion.