xAI launches Grok 4.6 with enhanced agentic capabilities

After unveiling Grok 4.5 in July, xAI presented Grok 4.6 on August 12, an evolution of its model focused on enhancing agentic capabilities. Beyond simple query responses, the model can follow a mission over time, analyze complex information, work on a codebase, or produce a complete deliverable. It is designed to remain operational during tasks with many steps, such as research, data analysis, software development, or document production, aiming to limit context breaks and let the AI continue long missions without constant user redirection. This direction is also reflected in tools around Grok, such as Grok Build, a programming agent that works interactively or in the background. Technically, Grok 4.6 has a context window of 500,000 tokens, accepts textual and visual inputs, and produces textual responses, with configurable reasoning effort levels. To position Grok 4.6 against rivals, xAI references the Artificial Analysis Intelligence Index, a composite of nine benchmarks. According to the company, Grok 4.6 scores 61 points, exceeding open-weight models like Kimi K3 and approaching top models like GPT-5.6 Sol Max, while Claude Opus 5 and Fable 5 still lead on specialized evaluations. xAI trained Grok 4.6 longer than Grok 4.5, using generated and selected data for reasoning, technical knowledge, and engineering, refined with supervised and reinforcement learning. The model is available via the xAI API as grok-4.6, also in Cursor and Grok Build.