The AI landscape continues its rapid growth with the launch of Claude Sonnet 4.5, Anthropic’s latest mainstream large language model. The company has made a bold claim, describing the new model as the “best coding model in the world.” One of its biggest highlights is its ability to code nonstop for up to 30 hours.
Claude Sonnet 4.5 arrives just months after its predecessor, Sonnet 4. However, it showcases a significant jump in capability, especially for developers. The model scores a reported 77.2% on the SWE-Bench Verified benchmark. The latter tests an AI’s ability to handle real-world GitHub pull requests. This score reportedly allows Sonnet 4.5 to outperform competing models from OpenAI and Google in coding tasks.
Furthermore, the model now leads the OSWorld benchmark—a test measuring real-world computer use tasks—with a success rate of 61.4%. This is a substantial increase over earlier versions of Claude.
Anthropic’s Claude Sonnet 4.5: The agent of endurance
What truly separates Sonnet 4.5 from its competition and previous Claude models is its capacity for sustained work. Anthropic reports that the model can now run autonomously for up to 30 hours, maintaining focus and performance throughout. This is a dramatic increase from the seven-hour limit seen in the previous flagship, Claude Opus 4.
This endurance transforms the model from a simple assistant into a capable agent. During early trials, Sonnet 4.5 reportedly demonstrated the ability to do more than just write an application. It could also execute complex, multi-step projects like deploying database services. It registered domain names and even performed SOC 2 security audits—all without human oversight.
To support this shift toward autonomy, Anthropic has given the model access to new features. The list includes virtual machines and memory and better context management for long-running processes.
An ecosystem for AI agents
Beyond the core model update, Anthropic introduced several tools designed to empower developers using Claude:
Claude Code Updates: Anthropic’s dedicated coding agent receives the Sonnet 4.5 model. New features include a Visual Studio Code extension for viewing real-time changes, improved status visibility in the terminal, and checkpoints that allow users to easily roll back code changes if the model makes errors.
Claude Agent SDK: Developers can now build their own custom AI agents using the same core infrastructure that powers Claude Code. The SDK includes tools for agent orchestration, memory, and managing context over extended tasks.
Imagine With Claude: Anthropic launched this temporary, high-end experiment to showcase the model’s capabilities. It allows Max subscribers to interact with Claude as it generates software and user interfaces on the fly, with no prewritten code or predetermined functionality.
Anthropic maintains that Sonnet 4.5 is its “most aligned” model to date. It features major safety improvements designed to resist prompt injection attacks and reduce concerning behaviors like sycophancy. Sonnet 4.5 is accessible through the Claude API and the claude.ai web app, with pricing remaining consistent with the previous Sonnet 4 model.
The post Anthropic’s Claude Sonnet 4.5 Can Code for Up To 30 Hours Straight appeared first on Android Headlines.
