Grok 4.6 focuses on staying with a complex task

The Grok team introduced Grok 4.6, emphasizing long-running agents and visual or interactive work. The announcement describes additional training and examples of research, codebase work, and application building, with initial availability in Cursor and Grok Build.

Context

Staying active for more steps is useful only when those steps move toward the requested outcome. A long task should remain inspectable: assumptions, tests, and unfinished work need to be visible. Selected project demonstrations can suggest what to try, while representative evaluations establish how consistently the capability appears.