Claude Sonnet 4.5
Frontier agents and coding — 30+ hours of focused accuracy.
Claude Sonnet 4.5, released in September 2025, is Anthropic's frontier model optimized for agents, coding, and computer use. It can work on complex coding projects for over 30 hours while maintaining accuracy and focus, achieving 77.2% on SWE-bench Verified. It supports a 200K token context window with up to 64K output tokens.
Performance Scores
Claude Sonnet 4.5 — Capability Radar
Strengths
77.2% on SWE-bench Verified — top-tier coding
30+ hour sustained coding accuracy
Excellent agentic and computer-use capabilities
200K context window with 64K output
Weaknesses
Premium pricing at $3/$15 per 1M tokens
Not as strong as Opus on deep philosophical reasoning
Speed slower than Haiku variants
Use Cases
Long-running agentic coding projects
Complex software development workflows
Computer use and UI automation
Large-scale code refactoring
Example Demo
Example Prompts
Large-scale refactoring
“Refactor this entire microservices codebase from REST to gRPC, maintaining backward compatibility.”
DevOps automation
“Build a complete CI/CD pipeline for this monorepo with automated testing and deployment.”
Computer use
“Navigate to this website, fill out the form, download the report, and summarize the key findings.”