Anthropic today released Claude Sonnet 4.5, its newest and most capable model, which the company claims is the best in the world for coding and building complex agents. The model arrives alongside a set of product upgrades including Claude Code, which introduces checkpoints for saving progress and rollback, a refreshed terminal interface, and a native VS Code extension. API pricing remains unchanged at $3 per million input tokens and $15 per million output tokens.

According to Anthropic, Sonnet 4.5 achieves state-of-the-art results on SWE-bench Verified with a score of 77.2% using a simple scaffold, and 82.0% with additional complexity and parallel test-time compute. The model also leads on the OSWorld benchmark for computer use at 61.4%, up from 42.2% four months ago. Anthropic researcher David Hershey told TechCrunch that in early enterprise trials he saw Sonnet 4.5 code autonomously for up to 30 hours — building an application, standing up database services, purchasing domain names, and running a SOC 2 security audit.

Alongside the model, Anthropic released the Claude Agent SDK, the same infrastructure that powers Claude Code, giving developers the tools to build their own agents. The company also launched a temporary research preview called Imagine with Claude, available to Max subscribers for five days, which demonstrates the model generating software on the fly in real time. Anthropic claims Sonnet 4.5 is its most aligned frontier model yet, with reduced sycophancy, deception, and resistance to prompt injection, released under ASL-3 protections.

Reactions from early partners were swift. Cursor CEO Michael Truell called Sonnet 4.5 state-of-the-art for coding, especially on longer horizon tasks. Windsurf CEO Jeff Wang echoed that sentiment, saying the model represents a new generation of coding models. The launch comes less than two months after Anthropic’s previous model, Claude Opus 4.1, in a race where leadership shifts every few months.