Anthropic's Claude Sonnet 4.5 Sets New Standard for Coding and AI Agents
Claude Sonnet 4.5 is now available worldwide via the Claude API.

Anthropic has unveiled Claude Sonnet 4.5, calling it the most advanced coding and agent-building model in the world. Positioned as the strongest model for computer use, reasoning, and math, it delivers a leap forward in real-world performance benchmarks and developer tools.
"This is the most aligned frontier model we’ve ever released, showing large improvements across several areas of alignment compared to previous Claude models," the startup said.
At its core, Sonnet 4.5 is designed to handle complex, multi-step coding tasks with unprecedented focus. On SWE-bench Verified, a benchmark for real-world software development, it leads all models.
On OSWorld, which measures AI performance in real-world computer tasks, Sonnet 4.5 posted a 61.4% score—up from 42.2% just four months ago. Anthropic says the model can stay on track for over 30 hours of continuous work on demanding tasks.
Introducing Claude Sonnet 4.5—the best coding model in the world.
— Claude (@claudeai) September 29, 2025
It's the strongest model for building complex agents. It's the best model at using computers. And it shows substantial gains on tests of reasoning and math. pic.twitter.com/7LwV9WPNAv
Alongside the release, Anthropic has rolled out major product upgrades. Claude Code now includes checkpoints to roll back progress, a redesigned terminal, and a native VS Code extension.
The Claude API gains context editing and extended memory, enabling longer-running agents, while the Claude app now supports direct code execution and file creation for spreadsheets, slides, and documents. The Claude for Chrome extension is rolling out to Max users.
For developers, Anthropic introduced the Claude Agent SDK, the same infrastructure behind Claude Code, offering a foundation to build powerful custom AI agents.
Beyond raw power, Sonnet 4.5 is Anthropic’s “most aligned model yet.” It features stronger defenses against prompt injection attacks, improvements in safety across high-risk domains like CBRN, and reduced tendencies toward sycophancy, deception, or harmful compliance.
Released under AI Safety Level 3 protections, the model balances capability with safeguards, including advanced classifiers to minimize risks.
With pricing unchanged from Sonnet 4 at $3/$15 per million tokens, Claude Sonnet 4.5 is now available worldwide via the Claude API, positioning itself as a frontier tool for developers, enterprises, and researchers alike.
Comments ()