Introducing Claude Opus 4.5

• Anthropic claims Opus 4.5 is best model globally for coding and agents.
• Scores higher than any human on Anthropic’s internal perf-engineering exam.
  • 2-hour window, real software engineering tasks.
  • 10.6% over Sonnet 4.5 on Aider Polyglot; 29% on Vending-Bench.
• Uses 65% fewer tokens on long-horizon coding vs prior Opus.
  • Medium effort setting matches Sonnet 4.5 at 76% fewer output tokens.
• “Effort parameter” lets devs trade capability for token efficiency.
• Most robustly aligned Anthropic model; best prompt-injection resistance yet.
  • $5/$25 per million tokens input/output.
• Claude Code gets Plan Mode with editable files; desktop adds parallel sessions.

· ** · Read on anthropic.com