Fable Brain

No Public Fable 5 Benchmark Data

3 min read

No Public Fable 5 Benchmark Data

Resolved 2026-07-07: this was a true gap before the Fable/Mythos system card was integrated, but it is no longer correct to say the vault has no public benchmark or evaluation data for Fable 5.

Done

The 319-page Claude Fable 5 and Claude Mythos 5 system card is now captured in Fable Mythos 5 System Card and the source ledger. It supplies public evaluation evidence for cyber capability, safeguards, prompt injection behavior, wellbeing, evenhandedness, child safety, and biosecurity threshold analysis.

What changed

  • The old gap asserted that every capability claim rested on launch framing or practitioner anecdote.
  • The system card now provides dated official evidence for the shared Fable/Mythos underlying model, including public cyber and safety evaluations.
  • The most important cyber figures now captured in the claim ledger are Mythos 5 ExploitBench mean score 10.75, CyberGym target reproduction 83.8 percent, Firefox 147 working exploit success 88.4 percent, and Fable safeguards flagging 407 of 410 ExploitBench episodes.
  • The system card also records that Anthropic classified Mythos 5 as CB-1 and Cyber Tier 1, while stating that the CB-2 judgment is less clear than for prior models and that the model can significantly uplift well-resourced threat actors.

Remaining caution

This gap is resolved, but the evidence is uneven. The public record is strongest for safety, dual-use, and red-team evaluations. General-purpose public benchmark tables for everyday coding, reasoning, vision, and agentic productivity remain thinner than the safety-card evidence.

Best practice

  • Use Fable Mythos 5 System Card as the primary source for public evaluation data. EVIDENCE-BASED
  • Stop saying “no public Fable 5 benchmark data”; say “public general-purpose benchmark coverage is thinner than safety and dual-use evaluation coverage.” EVIDENCE-BASED
  • Keep practitioner impressions separate from system-card evidence when ranking Fable 5 against Opus 4.8, Sonnet 5, or competitor models. PRACTITIONER
  • Before spending heavily on Fable 5, still run task-level evals for the actual workload. PRACTITIONER

Pitfalls

  • Treating a resolved gap as still active and downgrading every Fable 5 capability claim unnecessarily.
  • Treating safety and cyber evals as a full substitute for product-specific coding, writing, or agentic evals.
  • Ignoring that Fable 5 and Mythos 5 share the same underlying model, while Fable 5 includes safeguards that change some observable behavior.

Sources

Next actions

  • Keep this file as a resolved-gap audit trail until no active notes link to it as an open gap.
  • Watch for broader general-purpose benchmark tables and route them into Fable Mythos 5 System Card.
  • Retag downstream notes that still imply zero public evaluation data.

Graph View

Backlinks