Anthropic cuts Claude Code prompts by 80%, but Opus 5 adds 72% back
量子位 · wechat · 2026-07-27
Anthropic says it cut more than 80% of Claude Code’s old system prompt rules as it shifted from an oversized “employee handbook” to lighter context engineering. But a community trace of the actual prompts sent to different models found a more nuanced story: Opus 4.7 had 15,225 characters, Opus 4.8 dropped to 4,467, and Opus 5 climbed back to 7,694—about 72% longer than 4.8, though still roughly half of 4.7.
The article explains why both numbers can be true:
- The big reduction happened in the transition to the new prompt system for newer models.
- Opus 5 got extra instructions back because it behaves more proactively, uses sub-agents more often, and tends to produce longer outputs.
- The new additions mainly cover delivery/workflow control and correction behavior: scope management, progress reporting, when to stop, and how much to self-correct.
For developers using Claude Code or building agents on Claude, Anthropic’s guidance is to keep CLAUDE.md lean, split long instructions into Skills, remove duplicate rules, reduce tool-call examples, and provide executable references like tests or existing code instead of verbose prose.
Related event: Anthropic Slashes Claude Code System Prompts by 80%(3 posts)→
More from coding & agent
- theo builds his own visualizer for today's agent models, showing how cheap Luna really is — ivan_bezdomny · 2026-09-23
- Vite+ Hits RC: One Rust-Powered CLI to Replace Your Entire Web Toolchain — cnakazawa · 2026-09-23
- Tesla's in-car Grok agent books trips across Gmail, Calendar and Notion in one command — xiaohu · 2026-09-23
- Tesla's In-Car Grok Assistant Now Executes Cross-App Tasks in One Sentence — xiaohu · 2026-09-23
- Garry Tan says Capy lets him ship PRs much faster than Codex or Claude Code — garrytan · 2026-09-23
- DeskPilot: open-source native Python desktop client for local LLMs with MCP and sandboxed tools — poofph · 2026-09-23