Claude Sonnet 4.5 Summary
Overview
Claude Sonnet 4.5 is introduced as the world’s leading coding model, excelling at building complex agents, software coding, and computer usage. It brings improved reasoning, math capabilities, and substantial upgrades across Anthropic’s product suite.
Key Features and Improvements
- Best-in-class Coding: State-of-the-art performance on SWE-bench Verified, excelling at real-world software tasks.
- Powerful Agent SDK: The same infrastructure behind Claude Code is now available to developers for agent building.
- Product Upgrades:
- Code checkpoints for instant rollback.
- Refreshed terminal interface and VS Code extension.
- New context editing and memory features for longer, complex tasks.
- Support for code execution and file creation in chats.
- Chrome extension for Max users.
- Enhanced Multi-step Reasoning: Maintains focus over 30+ hours on multi-step, long-horizon tasks.
- Efficiency Improvements: Increased parallel tool use, better code test automation, and higher accuracy benchmarks.
Real-world Impact
- Significant productivity gains for developers and companies (e.g., Canva, Figma, Cursor, Devin).
- Improved coding accuracy, reduced error rates, and better codebase reasoning.
- Elevated security analysis and complex legal/financial task proficiency.
Alignment and Safety
- Most aligned model yet with advanced safety training and reduced risky behaviors (sycophancy, deception, prompt injection).
- Released under AI Safety Level 3 (ASL-3) with robust input/output filters for critical safety domains.
Access and Pricing
- Available immediately everywhere via the Claude API (
claude-sonnet-4-5). - Pricing matches Sonnet 4 at $3/$15 per million tokens.
Bonus Preview: Imagine with Claude
- Temporary research preview for Max subscribers lets users generate software on the fly using the model’s full agentic power.
Anthropic recommends all users upgrade to Claude Sonnet 4.5 for improved performance at no additional cost.