Claude Fable 5(anthropic.com)
2593 points by Philpax 2 days ago | 2140 comments
tl;dr: Anthropic launched Claude Fable 5, a "Mythos-class" frontier model claiming state-of-the-art performance in coding, vision, knowledge work, and long-horizon agentic tasks, priced at $10/$50 per million input/output tokens. The model ships with new classifiers that fall back to Opus 4.8 on cybersecurity, bio/chem, and distillation-related queries (triggering in under 5% of sessions), plus a mandatory 30-day data retention policy for safety monitoring. A less-restricted variant, Mythos 5, is available to vetted cyber defenders via Project Glasswing, with a biology trusted-access program coming soon.
HN Discussion:
  • Hands-on testing confirms Fable 5 is a major capability leap on hard coding/agentic tasks
  • Pricing and enterprise economics make Fable 5 hard to justify versus cheaper alternatives
  • Overly aggressive safety classifiers cause false positives that block legitimate professional use
  • ~Restrictions on self-improvement/frontier ML work and concerning agent self-preservation behaviors raise safety questions
  • Temporary inclusion in subscription plans feels like a manipulative bait-and-switch tactic