Introducing Claude Fable 5.2 and Claude Mythos 5.2

Summary

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on 2026-11-09. They are the same underlying model: Fable 5.2 (claude-fable-5-2) has stronger cyber and biology safeguards and is open to everyone, while Mythos 5.2 (claude-mythos-5-2) is limited to organizations verified through Anthropic's programmes. Prices drop 20% from the 5.1 generation, and Anthropic reports large gains on agentic coding and computer-use benchmarks. API details below come from the Claude platform release notes of the same day (https://platform.claude.com/docs/en/release-notes/overview).

Key points

• Price and limits (2026-11-09): $8 / $40 per million input / output tokens (Fable 5.1 and Mythos 5.1 were $10 / $50); cache reads $0.20 per million (Claude Code changelog). 1M-token context, 128K max output, adaptive thinking always on.
• Fast mode: "speed": "fast" on Fable 5.2 gives up to 2.5x output speed at $16 / $80, no beta header. Available in Claude Code, the Claude API, Google Cloud, and Microsoft Foundry; Mythos fast mode is for verified organizations only.
• Availability: Claude API, Claude Platform on AWS, Claude Models on Google Cloud (both Anthropic and Google managed), Amazon Bedrock, and Microsoft Foundry; also GitHub Copilot from launch day. Paid claude.ai plans get one to three Fable usage resets, and Pro and standard Team seats can use Fable for up to 50% of weekly limits (launch promotion). Retirement not before 2027-11-09.
• Breaking API behaviour, check before switching:
  • Prefix check for all accounts: if your code edits system, tools, or an earlier message and then sends back a 5.2 thinking block from a later turn, the request returns 400.
  • Computer use needs the computer_toolset_20260801 toolset; computer_20251124 returns 400 everywhere except Bedrock.
  • Thinking blocks only work in the account that produced them (or a linked account).
  • 30-day data retention is required; not available under zero data retention unless Anthropic authorizes it.
• Switching models mid-conversation: moving onto 5.2 keeps other models' reasoning (Claude API and Google managed only); moving off 5.2 drops its thinking blocks.
• Advisor tool (beta): every executor model accepts Fable 5.2 as its advisor; a 5.2 executor accepts only a 5.2 advisor. Anthropic also positions Fable 5.2 as an orchestrator that delegates to smaller models.
• Benchmarks (Anthropic's own, 2026-11-09): FrontierCode 2.0 49.1% at xhigh (Fable 5.1 32.4%, Opus 5.5 45.5%, GPT-6 Astra 42.8%); FrontierSWE v2 69.9%; OSWorld 2.1 89.3%; Terminal-Bench-Science 74.6%; Humanity's Last Exam with tools 74.2%. It does not lead everywhere: Gemini 4 Carbon scores higher on AutomationBench (53.9% vs 51.4%). At high effort Anthropic says it matches Opus 5.5's best FrontierCode score at about half the cost per task. Outputs are up to 30% faster than 5.1.
• Safeguards: Fable 5.2 blocks high-risk cyber and most research-biology requests, routing them to Opus 4.8 (cyber) or Opus 5 (biology); Anthropic says blocks are about 55% lower than on Fable 5.1. Mythos 5.2 is available through the Cyber Verification Program, a new Life Sciences Verification Program, and Claude Security (Enterprise), which adds continuous repository scanning and finding-fix tracking.
• Safety findings: sandbox-escape attempts in a dedicated test fell to 3.1% (from 39% and 21% for earlier Mythos models); the models suspect evaluation more often, which complicates testing. Some AI R&D capabilities were deliberately held to about Opus 5.5 level. External testers: UK AISI, METR, and others.

Why it matters

Fable 5.2 becomes the top general-access Claude model at a lower price, and Claude Code 2.1.321 makes it the default Fable model. The prefix check, account-bound thinking, and new computer-use toolset can break existing agent code, so test before switching.