Claude: System Prompts
Points and comments are a snapshot, not live.
Anthropic publishes the evolving system prompts used by Claude models across web and mobile apps.
Claude's web interface and mobile apps use system prompts to provide up-to-date information and encourage certain behaviors. These prompts are periodically updated to improve responses. The page lists prompts for many models, including Claude Opus 5, Fable 5, Sonnet 4, and earlier versions. Early prompts were around 300 words; later ones exceed 3,000 words. Starting with Claude 4.6, each model ID is a fixed snapshot with one entry. These updates do not apply to the Claude API.
What commenters are saying
Commenters note that early system prompts were around 300 words while the latest exceed 3,000, with one suggesting they accrete like building codes "written in blood." A user maintains a git history tracking changes. The Opus 5 prompt includes instructions about handling requests redirected from Fable 5 due to safeguards. Some argue lengthy prompts waste context tokens and may hurt performance. Others defend their value for complex tasks. Users point out that Anthropic omits tool definitions and Claude Code prompts, which can be extracted via a logging proxy.