Anthropic launches Claude Opus 5 at $5/$25 and says it tops coding benchmarks
APPSO · wechat · 2026-07-25
Anthropic has released Claude Opus 5, pricing it at $5 per million input tokens and $25 per million output tokens — the same as Opus 4.8 — while pushing the model to the top of several coding and knowledge-work benchmarks.
The article says Opus 5 performs especially well on software engineering tasks, where it can come close to flagship-level results at roughly half the cost in some benchmarks. It also improves on complex reasoning, autonomous task completion, and scientific evaluation suites, while Anthropic emphasizes stronger safety behavior, fewer deceptive actions, and more conservative handling of risky requests.
A notable workflow change is that Claude Code now favors much shorter system prompts. Anthropic says large parts of the old prompting rules were unnecessary, and that better tool interfaces plus progressive disclosure work better than heavy-handed instructions. Two beta updates also ship with the release: dynamic tool changes during a conversation without invalidating prompt cache, and an automatic fallback path for safety-blocked requests.
Related event: Anthropic Releases Claude Opus 5: New SOTA Performance at Half the Price(90 posts)→
More from coding & agent
- Claude Code 2.1.220 is reportedly about to be released — ClaudeCodeLog · 2026-07-25
- Fable helps sketch a latent-space game for Opus 5 — repligate · 2026-07-25
- Seroter’s July 24 reading list covers agentic specs, AI ROI data and brittle architecture — rseroter · 2026-07-25
- LM Studio’s Bionic launches as a local-first agent for docs, coding and voice — nicolascraske · 2026-07-25
- 'Clean Code' Author Rebuts: True Engineering Beats Vibecoding — burny_tech · 2026-07-25
- AI Security Agent Achieves RCE on GitLab Default Configuration via Dependency Chain — andreamichi · 2026-07-25