First hybrid reasoning model released February 2025, combining instant responses with visible extended thinking. Achieves state-of-the-art performance on SWE-bench and TAU-bench with significant improvements in coding and front-end development.
- First hybrid reasoning model integrating instant responses with extended thinking in one unified system
- State-of-the-art coding performance with 70.3% on SWE-bench Verified and exceptional frontend development
- Visible thought process in raw form with controllable thinking budgets up to 128K tokens
- 45% reduction in unnecessary refusals with improved understanding of harmful vs benign requests
- Enhanced agentic capabilities with improved computer use and multi-step task completion
Web Search
Search the web for current info.
Extended Thinking
Enhanced reasoning for complex tasks.
Model Information
Supported Formats
Claude Sonnet 4.5State-of-the-art coding model released September 2025, achieving 77.2% on SWE-bench Verified (82% with high compute). Excels at autonomous multi-step tasks for 30+ hours with enhanced tool coordination, context management, and computer use capabilities. Most aligned frontier model with improved security against prompt injection.
Claude Haiku 4.5Fast, cost-efficient model released October 2025, achieving 73.3% on SWE-bench Verified. Delivers performance comparable to Claude Sonnet 4 at one-third the cost and more than twice the speed. First Haiku model with extended thinking, computer use, and context awareness capabilities. ASL-2 safety classification with lower misalignment rates than larger models.
Claude Opus 4.5Premium model released November 2025, combining maximum intelligence with practical performance. Features 200K token context window with 64K max output, extended thinking support, and priority tier access. Knowledge cutoff March 2025 with training data through August 2025. Offers the best balance of intelligence and efficiency for complex reasoning tasks.