Skip to main content

Claude Sonnet 4.5

Claude Sonnet 4.5, released September 29, 2025, is Anthropic's "Anthropic's best coding model" with breakthrough advances in computer use, reasoning, and sustained focus. It maintains the same competitive pricing as Claude Sonnet 4 while delivering significantly enhanced capabilities.

Core Capabilities​

  • Anthropic's best coding model: State-of-the-art performance on SWE-bench Verified evaluation with superior code generation, debugging, and refactoring
  • Enhanced computer use: 61.4% on OSWorld benchmark (up from 42.2%), strongest model for building agents that interact with computer interfaces
  • Extended focus: Maintains concentration for 30+ hours on complex, multi-step tasks without losing context or quality
  • Superior reasoning: Significant gains in mathematical performance and logical analysis across finance, law, medicine, and STEM domains
  • Improved alignment: Most aligned frontier model with reduced sycophancy, deception, and better prompt injection defense

Technical Specifications​

  • Model ID: claude-sonnet-4-5-20250929
  • Pricing: $3/$15 per million tokens (input/output)
  • Context Window: 200K tokens standard, 1M tokens via API
  • Release Date: September 29, 2025

New Claude Code Features​

  • Checkpointing system: Automatically saves code state, instant reversion with Esc twice or /rewind command
  • Refreshed terminal: Improved status visibility and searchable prompt history (Ctrl+r)
  • VS Code extension: Beta extension with real-time change tracking and inline diff visualization
  • Context management: Automatic context editing and memory tools for optimized window usage and persistent knowledge
  • Code execution: Create and run code directly in Claude conversations

Performance Benchmarks​

  • SWE-bench Verified: State-of-the-art software engineering evaluation
  • OSWorld: 61.4% success rate (up from 42.2%)
  • Task endurance: 30+ hours sustained focus on complex workflows
  • Context editing: 29% performance boost, 84% token reduction in long sessions
  • Memory integration: 39% combined improvement with memory tools and context editing

Best Used For​

  • Complex development: Multi-phase implementations requiring extended focus and deep reasoning
  • Autonomous agents: Building agents that independently handle tasks with computer use capabilities
  • Code review: Architectural analysis and large-scale refactoring
  • Algorithm work: Advanced mathematical and logical problem-solving
  • System design: Complex architecture decisions requiring sustained reasoning

Usage​

# Set as default model
export ANTHROPIC_MODEL="claude-sonnet-4-5-20250929"

# Use for specific task
claude --model claude-sonnet-4-5-20250929 "Your task here"

# Or use /model command in Claude Code
/model

See Also: Claude News|Sonnet vs Opus|Change Model|Pricing

Last updated on