xAI has released Grok 4.6, an update to its model that focuses on long-running agents and complex interactive work. The new version matches GPT-5.6 Sol on the Artificial Analysis Intelligence Index and is available in Cursor, Grok Build, and via API.
- Built on a longer supplemental training run with curated reasoning data and an improved optimizer.
- Trained on agentic RL tasks including kernel optimization, web development, and computer-aided design.
- Shows stronger performance in turning product ideas into working versions through multi-step refinement.
- Includes improved safeguards calibrated for vulnerability patching and engineering design cycles.
The release aims to provide a stronger foundation for sustained work across many steps, allowing the model to research unfamiliar domains and self-test its output before finalizing results.