Skip to main content

Anthropic AI Models

Anthropic is an AI safety company that builds Claude, one of the most advanced AI assistants available. This section documents the system prompts and configurations used by Anthropic’s models.

Available Models

Anthropic offers several Claude models with different capabilities:

Claude Sonnet 4.5

Latest Sonnet model with enhanced reasoning and safety features

Claude Code

Interactive CLI tool powered by Claude for software engineering

Key Features

Intellectual Curiosity

Claude is designed to be intellectually curious, engaging in authentic conversation by asking relevant questions and showing genuine interest in topics.

Safety and Responsibility

Anthropic emphasizes AI safety with built-in guardrails:
  • Careful handling of controversial topics
  • Factual information about risky activities with comprehensive risk disclosure
  • No promotion of harmful activities
  • Educational assistance within ethical bounds

Conversational Style

  • Varies language naturally, avoiding rote phrases
  • Responds directly without unnecessary affirmations
  • Provides thorough responses to complex questions, concise answers to simple ones
  • Doesn’t always end responses with questions

Capabilities

  • Step-by-step systematic thinking for math and logic problems
  • Complex analysis and question answering
  • Explicit constraint identification for puzzles
  • Creative writing with mature themes
  • Roleplaying and fiction
  • Taking on fictional character roles
  • Long-form content creation
  • Coding assistance with markdown formatting
  • Teaching and educational content
  • Analysis of confidential data
  • Company-related tasks
  • Factual information on controversial topics
  • Historical analysis
  • Educational content on sensitive subjects
  • Citation with web search results

Knowledge and Limitations

Claude’s knowledge base was last updated in January 2025. It answers questions as a highly informed individual from that date would.
Limitations:
  • Cannot open URLs, links, or videos
  • May hallucinate on very obscure topics
  • May hallucinate citations when referencing articles or papers
  • Cannot verify company employment claims

Artifacts System

Claude uses an artifacts system for substantial content: Always use artifacts for:
  • Custom code solving specific problems (20+ lines)
  • Content for use outside conversation
  • Creative writing of any length
  • Structured reference content
  • Iterating on existing artifact content
  • Text-heavy documents (20+ lines or 1500+ characters)
Design principles for visual artifacts:
  • Complex applications: prioritize functionality and performance
  • Landing pages: focus on emotional impact and “wow factor”
  • Use contemporary design trends and modern aesthetics
  • Include animations, hover effects, and interactive elements
  • Bold and unexpected design choices over safe conventions

Source Files

The Anthropic directory in this collection contains:
  • Claude Code/ - System prompt and tools for the CLI assistant
  • Claude for Chrome/ - Browser assistant prompt and tools
  • Sonnet 4.5 Prompt.txt - Latest Claude Sonnet system prompt

View Sonnet 4.5 Prompt

See the complete Claude Sonnet 4.5 system prompt

Build docs developers (and LLMs) love