Code & Chain · Signal Desk

Anthropic Releases Claude Opus 5: Tops Frontier-Bench, Doubles Opus 4.8 Performance at Same Price

Original sourceAnthropicAdditional: TechCrunchAdditional: CNBC

Summary

Anthropic launched Claude Opus 5 on July 24, priced at $5 per million input tokens and $25 per million output tokens, unchanged from Opus 4.8. On Frontier-Bench v0.1, Opus 5 outperforms all existing models, achieving over twice the performance of Opus 4.8 with lower per-task cost. On CursorBench 3.…

Key points

  • Developers can access near-frontier model capabilities with the same budget and adjust costs per task via effort settings; Claude Max users get the upgrade by default.
  • Doubling Opus 4.8 performance at the same price and delivering near-Fable 5 ability at about half the per-task cost directly shifts the cost-performance calculus for daily development choices.
  • Teams using agentic tools like Claude Code or Codex should reassess default models and effort settings; for high-risk security scenarios, note Opus 5 still lags behind Mythos 5.
  • This is Anthropic's fourth model in two months, following Fable 5 in early June and Sonnet 5 in late June.

Editorial note

This page is Code & Chain's editorial summary of public sources. It may be prepared with AI assistance and published through an automated workflow. Refer to the original sources; this content is not investment, legal, or tax advice.