In a key benchmark, Claude Sonnet 4.5, a next-generation language model, runs autonomously for 30 hours on a single task.

The latest upgrade brings the ability to save your progress and create custom agents, with fewer behavioral issues, such as ...