Claude Sonnet 4.5 worked independently for 30 hours

Anthropic has unveiled its latest AI model, Claude Sonnet 4.5, which has shown remarkable progress in autonomous operation. The model ran autonomously for 30 hours, developing an entire chat software application similar to tools like Slack or Teams. According to the company, it created about 11,000 lines of code in the process, stopping only when it had fully completed the task.

The 30-hour battery life represents a significant leap forward from the previous model, the Opus 4, which lasted seven hours in May.

The achievement is a significant step in Anthropic’s battle to dominate the AI agent and AI coding market. The company called Claude Sonnet 4.5 “the world’s best model for real-world agents, coding, and computer use,” and emphasized that it leads the market in “computer use,” a reference to the Computer Use feature that Anthropic introduced nearly a year ago.

The model specializes in areas such as cybersecurity, financial services, and research. Canva was one of the first to test the new model, where they say the new model has helped them with “complex, long-context tasks, from engineering in our codebase to features within products and research.”

Anthropic also announced that the new model will come with additional updates to help developers code their own UI agents. This includes access to virtual machines, memory, context management, and multi-agent support. The company says that by doing so, it essentially provides the same building blocks that power Claude Code, allowing developers to create their own high-end agents.

Dianne Penn, head of product management at Anthropic, was surprised by the newfound power of their model, despite all the hype. Penn added that she uses the model herself to help with hiring, independently conducting in-depth web searches, setting parameters for profiles, and creating a spreadsheet of potential candidates and their LinkedIn profiles.