Claude Opus 5(anthropic.com)
1765 points by alvis 48 days ago | 1315 comments
tl;dr: Anthropic released Claude Opus 5, claiming state-of-the-art performance on coding and knowledge-work benchmarks (Frontier-Bench, GDPval-AA, ARC-AGI 3, OSWorld 2.0) at half the cost of their top-tier "Fable 5" model, though it still trails "Mythos 5" on cybersecurity exploit development and advanced biology tasks. Pricing matches Opus 4.8 at $5/$25 per million input/output tokens, and Anthropic is loosening cyber safeguards (blocking ~85% less often than Fable 5) while adding beta features like mid-conversation tool changes and automatic model fallbacks for flagged requests.
HN Discussion:
  • Awe at rapid pace of AI capability improvements becoming normalized
  • Data retention policy differences matter more than raw benchmark performance
  • ~Opus 5 retains annoying writing quirks unlike Fable 5
  • Anthropic's messaging is confusing and contradictory about Opus 5 vs Fable 5 capabilities
  • Skepticism about benchmark numbers discrepancy with original OSWorld authors