Anthropic has unveiled its newest AI model, Claude Sonnet 4.5, which was able to autonomously generate approximately 11,000 lines of code over a span of 30 hours. The project involved creating a chat application similar to popular platforms like Slack and Microsoft Teams, with the model continuing its execution until the task was fully accomplished.
This significant advancement contrasts with the company’s previous model, Opus 4, which generated considerable attention in May for completing tasks in a mere seven hours. The battery of improvements demonstrated by Claude Sonnet 4.5 marks a critical milestone for Anthropic in its quest to dominate the realms of AI agents and programming.
Anthropic described Claude Sonnet 4.5 as “the best model in the world for real-world agents, coding, and computer use,” emphasizing its capabilities in sectors such as cybersecurity, financial services, and research. One of the model’s early adopters, Canva, noted that it effectively managed “complex, long-context tasks—from engineering in our codebase to in-product features and research.”
Various tech companies, including OpenAI and Google, are persistently rolling out incremental updates to enhance the functionality of their AI systems. These advancements position their technologies as indispensable tools for both consumers—facilitating research and scheduling—and enterprises, aiding in presentations and data analysis. Just days prior, OpenAI introduced Pulse, a new ChatGPT feature tailored for daily routines and relevant research.
In conjunction with the model’s introduction, Anthropic plans to provide developers with updates that enhance their ability to create AI agents.
“We are merging the model launch with access to virtual machines, memory, context management, and multi-agent support,” the company stated in a release. “This effectively consolidates the fundamental components that drive Claude Code, allowing developers to construct their own state-of-the-art agents.”
Dianne Penn, Anthropic’s head of product management, expressed surprise over the enhancements in the model’s computer interaction capabilities. In an interview with Technology News, she noted that Claude Sonnet 4.5 is over three times more proficient than Anthropic’s technology from last October in navigating web browsers and utilizing various computing tools. Feedback from early-access users, including prominent figures from GitHub and Cursor, contributed to intensive refinements made to the model.
Scott White, the product lead for Claude.ai, informed Technology News that the model functions at a “chief-of-staff level.” It can coordinate schedules among multiple users, analyze data dashboards for insights, and prepare status updates based on individual meetings.
Although vibe-coding with the new model had not yet been tested by White or Penn at the time of the interview, Penn shared that she employs Claude Sonnet 4.5 in recruitment activities for new team members at Anthropic.
“Having a continuous running prompt for deep web searches, tailored to parameters for sourcing profiles for specific roles, has been tremendously beneficial,” Penn noted. “I’ve observed improvements in the quality and depth of searches, resulting in spreadsheets filled with LinkedIn profiles that I can then utilize for outreach.”
