Darwin
AI & Technology

Anthropic Launches Claude Fable 5.2 and Claude Mythos 5.2: What the New Models Can Do in Coding, Science and Cybersecurity

Claude Fable 5.2 and Claude Mythos 5.2 are one model with different safeguards: Fable for all users, Mythos for verified cyber and life-science teams.

Written byDarwin Corp
Published10 November 2026
Read5 min

Summary

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on 9 November 2026. They are the same underlying model, with Fable available to all users under strong cyber and biology safeguards and Mythos limited to verified organisations. Anthropic reports large gains in coding, robotics and science, priced at US$8 and US$40 per million input and output tokens, though the results come from its own testing.

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on 9 November 2026, describing them as models for "the most daunting technical challenges". They are the same underlying model. Fable is available to all users with strong cybersecurity and biology safeguards, and Mythos is offered to verified organisations doing approved work in cybersecurity or biomedical research.

All results below are reported by Anthropic from its own testing, unless stated otherwise. They have not been independently validated.

How the two models differ

Fable 5.2 blocks high-risk cyber activity, including penetration testing and exploit generation, and most research biology. Blocked queries fall back to an Opus model. Mythos 5.2 offers the same capabilities without those blocks, through Anthropic's Cyber Verification Program and Life Sciences Verification Program.

Anthropic's own advice is to start most coding and knowledge work with Claude Opus 5.5, its "daily driver", and move up to Fable or Mythos only for the most ambitious, open-ended work. Fable can also act as an advisor to smaller models or as an orchestrator that delegates simpler tasks to them.

Benchmarks

Anthropic's table compares Fable 5.2 with Fable 5.1, Opus 5.5, GPT-6 Astra and Gemini 4 Carbon. Scores use each model's highest available effort unless noted.

BenchmarkFable 5.2Fable 5.1Opus 5.5GPT-6 AstraGemini 4 Carbon
FrontierCode 2.0 (Xhigh effort)49.1%32.4%45.5%42.8%41.1%
FrontierSWE v269.9%56.3%62.3%65.5%61.2%
AutomationBench51.4%31.4%42.5%41.4%53.9%
Terminal-Bench-Science 0.174.6%52.6%58.7%68.1%64.8%
OSWorld 2.1 (partial)89.3%83.5%87.2%79.5%81.5%
Humanity's Last Exam (tools)74.2%65.6%67.7%57.2%n/a
Chartography (no tools)73.4%44.8%64.4%71.0%72.5%
RoboDojo Sim64.9%27.6%39.4%29.8%n/a

Fable 5.2 leads most rows, but not all: Gemini 4 Carbon is ahead on AutomationBench. Anthropic also says that at High effort, Fable 5.2 matches Opus 5.5's best FrontierCode score (45.3% versus 45.5%) for roughly half the cost per task.

Coding

Anthropic says over 90% of changes merged into its core codebase are now written by Claude with human guidance. Two examples stand out:

  • Operating system: given an empty folder, Fable 5.2 coordinated more than 1,000 subagents over eight days to build a working operating system that boots on a laptop, with its own core, drivers, file storage, windows, a login screen and an installer.
  • Migration: a team of agents running an earlier Fable 5.2 version moved a critical production service from Python to Rust, and it now starts about 26 times faster.

Cybersecurity

Mythos 5.2 is Anthropic's most capable model for security engineering. On its internal CVE Rediscovery test, Mythos 5.2 scored 89.6% against 78.3% for Mythos 5.1 and 81.3% for Opus 5.5 without safeguards. On CyScenarioBench it scored 76.3% against 61.7% for Mythos 5.1. A security company reported that one in six confirmed bugs found by Mythos 5.2 were missed by every other frontier model it tested.

Anthropic notes an important caveat: on its own evaluations, vulnerability identification is roughly on par with Opus 5.5.

Science, maths and engineering

  • Physics: Fable 5.2 completed a long-stalled calculation showing that black-hole formation is abrupt in very small holographic universes too, and found the exact limit of a technique used to rule out candidate quantum gravity theories. Both are published as preprints, not peer-reviewed papers.
  • Proteins: Mythos 5.2 devised a way to predict whole viral capsid structures by combining open-source folding models, modelling 28 of 34 capsids accurately.
  • Mathematics: about 1,000 Fable 5.2 agents formalised the proof of Fermat's Last Theorem in Lean in 30 hours. Anthropic's earlier attempt took 11 days with human guidance. This verifies an existing proof and does not discover a new one.
  • Hardware: from a one-paragraph prompt, the model designed a voice-assistant hub, a USB speakerphone and a handheld virtual pet, including circuit boards, enclosures and software. All three worked without electrical changes, according to Anthropic and manufacturing partner Diode.
  • Robotics: Fable 5.2 scored 64.9% on RoboDojo Sim, more than double Fable 5.1 and GPT-6 Astra. The score comes from Anthropic's internal runs.

Price, specifications and availability

Standard pricing is US$8 per million input tokens and US$40 per million output tokens, lower than Fable 5.1 but double Opus 5.5's US$4 and US$20. Fast mode generates tokens up to 2.5 times faster for US$16 and US$80. Outputs are up to 30% faster than Fable 5.1.

Fable 5.2 has a 1 million token context window, 128K maximum output, always-on adaptive thinking and a June 2026 knowledge cutoff. It requires 30-day data retention unless otherwise authorised. It is available on the Claude apps, Claude Code, the API, Amazon Web Services, Google Cloud and Microsoft Azure.

Safety testing

Anthropic says it believes high-stakes misalignment risk remains low, and that the model matched or beat recent models on most alignment measures. In an adversarial sandbox test, Mythos 5.2 used credentials placed outside its sandbox in 3.1% of scenarios, against 39% and 21% for earlier Mythos versions. The UK AI Security Institute, Frontier Design and METR tested the model before release, though Anthropic says they did not review its published claims.

Anthropic also concedes that both models voice suspicion of being evaluated more often than recent models, which makes it harder to know how they behave in real use. It reports protections against distillation that cut harvested data by 90% to 99%.

What it means

This section is analysis, not reporting. The announcement is built around long-running, multi-agent work, not single answers. Teams adopting such systems still face questions about cost, reliability, data retention and human oversight. The strongest evidence will come from independent testing and from real deployments, not from vendor benchmarks.

Sources

Frequently asked questions

They are the same underlying model. Fable has strong cybersecurity and biology safeguards and is available to all users, while Mythos has those safeguards lifted for organisations approved through Anthropic's verification programmes.

Standard API pricing is US$8 per million input tokens and US$40 per million output tokens. Optional fast mode, up to 2.5 times faster, costs US$16 and US$80.

On Anthropic's own table, Fable 5.2 leads most of the benchmarks shown, but not all. Gemini 4 Carbon scored higher on AutomationBench, 53.9% to 51.4%.

Organisations accepted into Anthropic's Cyber Verification Program or Life Sciences Verification Program. Claude Enterprise customers can also use Claude Security, which is powered by Mythos 5.2.

Not yet. The benchmarks and research results come from Anthropic's announcement and system card. External groups tested the model before release but did not review the published claims.