xAI has introduced Grok 4.6, the newest Grok model — and the announcement itself is a story: it landed on Cursor’s blog, the agentic coding editor acquired by SpaceX days earlier.
Long-Running Agents
Grok 4.6 is tuned for the two hardest requirements of agentic coding: long-running agents that stay coherent across hours of work, and more ambitious interactive and visual work — reading UIs, interpreting screenshots, acting on what they see.
The launch continues a busy Grok cadence. Grok 4.5 delivered Opus-class performance at half competitor pricing; Grok Imagine video beat Sora, Veo, and Kling in blind user benchmarks; and xAI has been under scrutiny in parallel for its handling of deepfake misuse lawsuits.
Read the announcement at cursor.com/blog/grok-4-6.