Unlock Claude Sonnet 4.5: Your Next Coding Breakthrough
▼ Summary
– Anthropic released Claude Sonnet 4.5, claiming it is the best coding model in the world and excels in complex agent tasks, computer use, reasoning, and math.
– The model outperformed previous versions and competitors on the SWE-bench Verified coding benchmark and achieved a leading 61.4% score on the OSWorld benchmark for computer tasks.
– Claude Sonnet 4.5 is Anthropic’s most aligned model yet, adhering closely to human instructions with reduced sycophancy and deception and improved resistance to prompt injection attacks.
– It is available everywhere, including in the Claude.ai chatbot and API, at the same price as Sonnet 4, with upgrades to Claude Code like checkpoints and a refreshed terminal interface.
– Anthropic launched the Claude Agent SDK for developers to build their own agents and enhanced the Claude Code API with context editing and memory tools for tackling complex problems.
Anthropic has officially launched Claude Sonnet 4.5, positioning it as the premier coding model currently available. This latest iteration builds upon the solid foundation laid by its predecessor, Claude 4 Sonnet, delivering substantial performance gains across coding, reasoning, and computer interaction tasks. Developers now have access to a more powerful and refined toolset designed to accelerate software development workflows.
According to Anthropic, Claude Sonnet 4.5 has demonstrated exceptional results on the SWE-bench Verified evaluation, a carefully curated benchmark for assessing real-world software engineering capabilities. The model not only surpassed its own previous versions but also outperformed competing models from other leading AI labs. A particularly notable achievement is its ability to maintain focus on intricate, multi-step tasks for durations exceeding thirty hours, a trait that proves invaluable for autonomous agentic operations running in the background.
Beyond coding, the model shows marked improvement in computer task performance, achieving a top score of 61.4% on the OSWorld benchmark. This represents a significant leap from the 42.2% score recorded by Sonnet 4 just a few months prior. These enhanced capabilities are integrated into the widely available Claude for Chrome extension, which recently exited its waitlist phase. The model also exhibits stronger mathematical and logical reasoning skills.
Anthropic emphasizes that Claude Sonnet 4.5 is their most aligned frontier model to date. It demonstrates a higher degree of adherence to user instructions and intended applications while reducing undesirable behaviors. The model incorporates advanced safety features, including improved resistance to prompt injection attacks and AI Safety Level 3 protections within Anthropic’s framework.
Accessibility remains straightforward. The new model is available through the standard Claude.ai chatbot interface. Developers can integrate it via the API and within Claude Code, with pricing consistent with the previous Sonnet 4 model.
The launch includes substantial upgrades to Anthropic’s coding ecosystem. Claude Code now features progress checkpoints, allowing users to save their work and return to earlier states effortlessly. The terminal interface has been refreshed, and a native Visual Studio Code extension is now part of the package.
Furthermore, Anthropic released the Claude Agent SDK, providing developers with the same underlying infrastructure that powers Claude Code. This enables the creation of custom AI agents. The Claude Code API introduces a new context editing feature and a memory tool, empowering agents to operate more efficiently and handle increasingly complex challenges. Claude applications have also been enhanced with the ability to execute code and generate files directly within chat sessions, streamlining the development process even further.
(Source: ZDNET)
