Wednesday Jun 10
PRODUCTAnthropic

Anthropic Opens Mythos to the Public

10JUN
80.3% SWEGLASSWINGAPI NOW

Anthropic's most powerful model just left the lab. Claude Fable 5, the first public Mythos model, scores 80.3% on SWE-Bench Pro. Six weeks after it was reserved for cyber defenders, it's live in the API.

The Mythos model class launched in April, government-only. Yesterday, Fable 5 - the same intelligence - opened to everyone via API.

Stripe ran Fable 5 on a 50-million-line Ruby codebase. What took a team two months, Fable finished in one day. New safety classifiers fall back to Opus 4.8 in under 5% of sessions.

Pricing: $10/$50 per million tokens - double Opus 4.8. Mandatory 30-day data retention now applies - even if your enterprise contract said zero.

full brief & sources

⚡ Why this matters

  • Mythos-class capability moving from government-only to public API changes what any developer can build - and what attackers can access
  • The 5% safety fallback is a live experiment in whether guarded access can replace restricted access at frontier capability levels
  • Mandatory 30-day retention overrides existing enterprise zero-retention contracts - legal and compliance teams need to know this now

🔍 What happened

  • Anthropic opened Claude Fable 5 to the public API on June 9, 2026 - six weeks after Mythos-class models launched as government-only
  • SWE-Bench Pro score: 80.3%, the highest publicly reported score to date
  • Stripe completed a 50M-line Ruby codebase migration in one day - a task their team estimated at two months
  • Safety classifiers automatically fall back to Opus 4.8 in under 5% of sessions when high-risk patterns are detected

💬 Smart takes

  • UK AISI's pre-release red-teaming found early progress toward a universal jailbreak in a small fraction of sessions
  • The pricing doubles Opus 4.8: $10 input / $50 output per million tokens
  • The mandatory retention clause overrides any existing enterprise zero-retention agreement - a legal exposure for regulated industries

🧭 Where this goes

  1. Anthropic newsroom, TechCrunch, CNBC - June 9 2026

🥄 The Spoon Take

Six weeks ago, this capability was classified - government-cleared defenders only. Now it's in the API at $10/MTok. The safety classifiers are the real product. Anthropic is betting guarded access beats no access at every tier. If it holds, this is the template for every dangerous-but-useful model release that follows.

🤔 Pushback

The 5% fallback rate is self-reported - and the UK AISI already found progress toward a universal jailbreak in early testing.

·Anthropic - Claude Fable 5 and Mythos 5·TechCrunch - First public Mythos model launch·CNBC - Fable 5 general availability