Claude Sonnet 4.5 autonomously generates Slack clone in one shot in 30 hours (www.theverge.com)

Anthropic today unveiled Claude Sonnet 4.5, a new model that ran autonomously for 30 hours to build a Slack/Teams-like chat app and produced roughly 11,000 lines of code — stopping only when the task was complete. That endurance marks a major step up from Anthropic’s Opus 4, which previously demonstrated about seven hours of continuous operation. Anthropic bills Sonnet 4.5 as “the best model in the world for real‑world agents, coding, and computer use,” and early users like Canva report it excels on complex, long‑context engineering and research tasks. Technically, Sonnet 4.5 improves the “Computer Use” capability substantially (Anthropic says it’s more than three times better than last October) and ships alongside tools for developers: virtual machines, memory, context management, and multi‑agent support — essentially the building blocks for custom AI agents. Anthropic positions the model for enterprise domains such as cybersecurity, finance, and research; product leads describe “chief‑of‑staff” abilities like scheduling across calendars, extracting insights from dashboards, drafting status updates, and sourcing candidate profiles into spreadsheets. The release tightens competition among Anthropic, OpenAI, Google and others over autonomous agents and AI coding, while lowering the bar for developers to build persistent, long‑running agent workflows.