Anthropic has unveiled Claude Sonnet 4.5, a significant advancement in artificial intelligence designed to enhance coding and reasoning tasks. This latest iteration boasts state-of-the-art performance, achieving a 77.2% success rate on the SWE-bench Verified evaluation, a benchmark for real-world software coding abilities. The model demonstrates an extended attention span, capable of maintaining focus for over 30 hours on complex, multi-step tasks, a crucial development for agentic applications that require sustained operation. This enhanced capability allows for more intricate problem-solving and development workflows, positioning Claude Sonnet 4.5 as a powerful tool for professionals. The release is accompanied by a suite of new tools and enhancements aimed at democratizing advanced AI.
Claude Code Checkpoints have been introduced, providing developers with the ability to save their progress and revert to previous states, a feature highly requested by users for managing complex coding projects. The Claude API has been upgraded with context editing and memory tools, enabling longer tasks and more efficient handling of information outside the traditional context window. Furthermore, Claude applications now feature integrated code execution and file creation capabilities, allowing users to directly run code and generate documents within the conversational interface. Anthropic is also empowering developers with the Claude Agent SDK, providing the same infrastructure used to build Claude Code. This SDK allows for the creation of sophisticated AI agents capable of managing memory, handling permissions, and coordinating sub-agents for complex problem-solving.
The Claude for Chrome extension, previously on a waitlist, is now available to Max users who joined the list last month, integrating Claude's capabilities directly into their web browsing experience. This extension allows Claude to navigate websites, interact with elements, and automate tasks, offering a more seamless and intelligent browsing experience. Claude Sonnet 4.5 is positioned as a direct competitor to leading AI offerings from OpenAI and Google, with Anthropic emphasizing its improved performance at the same price point as its predecessor, Sonnet 4. The model's performance on the OSWorld benchmark, which assesses real-world computer use, has reached 61.4%, a significant increase from previous versions. Anthropic highlights that Sonnet 4.5 is their "most aligned frontier model yet," with notable improvements in reducing undesirable behaviors such as sycophancy and deception, underscoring a commitment to safety and responsible AI development. The release signifies a pivotal moment in AI development, offering enhanced tools and capabilities that foster innovation and productivity across various domains.