ClaudeMap

Claude Fable 5.2 & Mythos 5.2 Announcement

Anthropic's announcement of Fable 5.2 and Mythos 5.2 — one base model, 1M context, lower prices, and an optional fast mode up to 2.5x.

Tutorials & Guidesofficialmodelsannouncementfablemythos

Anthropic announced Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026: one base model, two guardrail postures. Fable 5.2 (API id claude-fable-5-2, a 1M-token context window) targets the most ambitious open-ended technical work, priced at $8/M input and $40/M output — down from Fable 5.1's $10/$50 — with an optional fast mode (up to 2.5x generation, $16/$80), and it became the default Fable model as of Claude Code 2.1.321 ($0.20/M cache reads); Mythos 5.2 is the same base model with relaxed guardrails, still gated to the Cyber and Life Sciences Verification Programs. Cited results include FrontierCode 2.0 at 49.1% (Xhigh effort), FrontierSWE v2 69.9%, WANDR (relaxed) 82.1%, OSWorld 2.1 (offline set, partial) 89.3%, the Remote Labor Index at 41.3% (High effort), and HLE with tools 74.2%. Anthropic recommends starting with Opus 5.5 for most coding and knowledge work and stepping up to Fable or down to Sonnet 5.5/Haiku 5.5 as scope demands; Fable 5.2 can also serve as Advisor or Orchestrator.

What it does

  • Anthropic's official announcement for the Claude Fable 5.2 and Mythos 5.2 models (November 9, 2026)
  • One base model, two guardrail postures: Fable 5.2 ships to everyone with strong cybersecurity and biology safeguards; Mythos 5.2 is gated to the Cyber and Life Sciences Verification Programs
  • First-hand pricing and availability: $8/M input and $40/M output (down from $10/$50 on 5.1), an optional fast mode at $16/$80 with up to 2.5x generation, a 1M-token context window, and same-day availability on AWS, Google Cloud and Azure
  • Benchmark tables run with safeguards on: FrontierCode 2.0 49.1% at Xhigh effort, FrontierSWE v2 69.9%, OSWorld 2.1 (offline set, partial) 89.3%, Remote Labor Index 41.3% at High effort, HLE with tools 74.2%
  • Positioning guidance: start with Opus 5.5 for most coding and knowledge work, step up to Fable/Mythos for the most ambitious work; Fable 5.2 can act as Advisor or Orchestrator
  • Safety detail: strongest-ever anti-distillation protections (90-99% less data harvested), preserved thinking now on all accounts, and Enterprise Frontier Safeguards in private beta

When to use it

Read when you need Anthropic's own framing of the current flagship tier — what Fable/Mythos 5.2 are for, how they sit against Opus 5.5 and the smaller models, and what they cost — before picking a model for large, open-ended technical work.

Getting started

No install — an announcement page. Fable 5.2 is available on all platforms the day of announcement (Claude apps, the Claude API via claude-fable-5-2, Amazon Web Services, Google Cloud, Microsoft Azure); fast mode runs on Claude Code, the Claude Platform, Google Cloud (Anthropic managed and Google managed) and Microsoft Foundry. Mythos 5.2 requires applying through the Cyber or Life Sciences Verification Program, and Mythos fast mode is limited to Verification Program organizations.

Things to know

Marketing-adjacent source: capabilities and benchmark numbers are described by the vendor — verify independent benchmarks for hard numbers. The 30% faster outputs and the 2.5x fast-mode figure are Anthropic's own measurements. A launch promotion gives eligible paid plans one to three Fable resets (by subscription length) and Fable up to 50% of weekly usage limits on Pro plans and standard Team seats, which inflates short-term availability. Benchmark table qualifiers matter: OSWorld 2.1 is the offline set with partial scoring, and CyScenarioBench scores are based on each lab's unique tasks. The announcement's system-card link 307-redirects to a PDF hosted on assets.claude.com.

Charted

Related on the map

  • Anthropic's official "Code with Claude" workshop materials — nine hands-on courses on skills, MCP, evals and Claude Managed Agents.

    • official
    • workshops
    • anthropic
    • managed-agents
  • Anthropic's canonical essay on agent design — workflows vs agents, and five composable patterns.

    • official
    • agents
    • engineering
    • patterns