xAI has released Grok 4.6, a model built for long-running agents and ambitious coding, research, visual, and interactive work. It builds on Grok 4.5 by handling complex assignments across many steps, from analyzing information and navigating a codebase to turning a broad product idea into a working application or a polished artifact. The release targets developers and knowledge workers who need an agent to carry a project from initial research through implementation and revision.
Behind the model is a longer supplemental training run than Grok 4.5 received. xAI used curated model-generated reasoning and technical material, high-quality engineering data, and a revised optimizer and training recipe. Grok 4.5 then regenerated SFT trajectories across reasoning levels, agent harnesses, STEM, software engineering, and knowledge work. Reinforcement learning covered general coding, kernel optimization, web development, and computer-aided design environments.
xAI reports a score of 61 on the Artificial Analysis Intelligence Index, up from 56 for Grok 4.5 High and level with GPT-5.6 Sol Max, while Fable 5 Max scored 62. Grok 4.6 led the listed systems on GDPVal-AA v2 and AA-Briefcase, but trailed GPT-5.6 Sol Max and Fable 5 Max on DeepSWE and Terminal-Bench. That mixed result positions it near the frontier while showing that its strongest gains are not uniform across every coding task.
Beyond benchmark scores, xAI says the model produces stronger first passes for visual and interactive projects, can establish an application structure and visual language in one pass, and shows more self-testing on longer runs. It can research an unfamiliar domain, implement core functions, and keep refining the result through feedback, making sustained project work the central pitch rather than one-shot code generation.
Grok 4.6 is available now in Cursor, Grok Build, the xAI API, OpenRouter, Vercel, and Cloudflare. API pricing starts at $2 per million input tokens and $6 per million output tokens, while the fast variant costs twice as much. Cursor and Grok Build include double usage for the first week. xAI also says safeguards were calibrated to the model’s capabilities, backed by its widest pre-deployment test suite and continued post-deployment and third-party testing.