Today we are releasing Grok 4.6, focusing on long-running agents and interactive/visual work. It handles complex multi-step tasks like researching, analyzing, coding, and building applications. Grok 4.6 matches GPT-5.6 Sol on the Artificial Analysis Intelligence Index and scores 61 on AA Intelligence, 1753 on GDPVal-AA, and 69.9% on Cursor Bench v3.2. It outperforms previous versions on benchmarks like Deep SWE v1.1 (65.9%) and Frontier Code v1.1 (61.3%).

Available today in Cursor and Grok Build with 2x included usage for the first week. Training used curated model-generated data, improved optimizer, and RL across knowledge work, coding, and domain-specific environments. Grok 4.6 excels at turning ideas into working projects, self-testing, and visual/interactive first passes.

Safety safeguards improved and calibrated. Pricing starts at $2 per million input tokens and $6 per million output tokens. Available via API and partners like Open Router, Vercel, Cloudflare.