Anthropic Releases Claude Sonnet 4.5
Claude Sonnet 4.5

September 29, 2025 – Anthropic has launched Claude Sonnet 4.5, calling it the most powerful and aligned AI model the company has ever released. Positioned as the world’s best coding model, Claude Sonnet 4.5 delivers major advances in coding, reasoning, math, and real-world computer use.

Key Upgrades in Claude Sonnet 4.5

Claude Sonnet 4.5 isn’t just a model update—it comes with a wave of product improvements across the Claude ecosystem:

  • Claude Code Enhancements

    • Checkpoints: Developers can now save progress and instantly roll back to earlier states.

    • Refreshed terminal interface and a new native VS Code extension.

    • Context editing and memory tools in the Claude API for longer, more complex agent tasks.

  • Claude Apps

    • Support for code execution and file creation (spreadsheets, slides, and documents) directly in chat.

    • Expanded rollout of the Claude for Chrome extension to Max users.

  • Claude Agent SDK

    • Anthropic is opening up the infrastructure behind Claude Code to developers.

    • The SDK provides tools for building autonomous agents with advanced memory, permission systems, and subagent coordination.

Performance Benchmarks

Claude Sonnet 4.5 sets new records across industry-standard benchmarks:

  • SWE-bench Verified (software engineering tasks): Leads all models.

    6421e7049ff8b2c4591497ec92dc4157b2ac1b30-3840x2160 copy
  • OSWorld benchmark (real-world computer use): Scores 61.4%, a sharp jump from 42.2% with Sonnet 4 just four months ago.

  • Substantial improvements in reasoning, math, and domain expertise across finance, law, medicine, and STEM.

Early testing shows Claude Sonnet 4.5 can maintain focus for 30+ hours on complex, multi-step tasks.

2825bb34cb9b9aa1ef347f816c590b5bbdae2b4d-2600x2288 copy

Images Courtesy: Anthropic

Safety and Alignment

Claude Sonnet 4.5 is Anthropic’s most aligned model to date, showing large reductions in behaviors such as sycophancy, deception, and unsafe compliance. It is released under AI Safety Level 3 (ASL-3) protections, which include classifiers designed to detect dangerous use cases like CBRN (chemical, biological, radiological, nuclear) risks.

Imagine with Claude

Anthropic is also launching a short-term research demo, Imagine with Claude, which showcases the model’s ability to generate and adapt software in real time. The preview is available for Claude Max subscribers for the next five days.

Pricing and Availability

Claude Sonnet 4.5 is available globally today through:

  • Claude API (claude-sonnet-4-5)

  • Claude Code and the Claude apps

  • Claude Developer Platform, including the new Agent SDK

Pricing remains unchanged from Claude Sonnet 4 at $3/$15 per million tokens.