Key Features:
• Hybrid Reasoning
-1. Combines standard quick responses with extended thinking mode for step-by-step reasoning
-2. First Anthropic model to offer adjustable reasoning budget (up to 128K tokens in beta)
-3. Excels in math, physics, and instruction-following with extended thinking
• Coding Excellence
-1. State-of-the-art performance on SWE-bench Verified (70.3% with scaffolding)
-2. Supports agentic coding via Claude Code CLI tool (preview)
-3. Enhanced front-end web development and GitHub integration
• Extended Output Capacity
-1. Outputs up to 128K tokens in extended thinking mode (beta, requires special API header)
-2. Ideal for comprehensive content generation and complex workflows
• Performance Boost
-1. Outperforms Claude 3.5 Sonnet in reasoning, coding, and real-world tasks
-2. Reduced unnecessary refusals by 45% for better usability
• Accessibility
-1. Available on Claude.ai, Anthropic API, Amazon Bedrock, and Google Cloud Vertex AI
-2. Pricing: $3 per million input tokens, $15 per million output tokens
• Practical Applications
-1. Powers AI agents, data analysis, and customer-facing workflows
-2. Early web search prototype for up-to-date answers
