xAI has launched Grok 4.6, a new AI model designed to handle complex, multi-step tasks that require sustained work rather than a single response. Grok 4.6 is built with AI agents in mind. It can work with unfamiliar codebases, build applications, analyze data, conduct research, and move through multiple stages of a project toward a defined goal.
One of the model’s key improvements is its ability to check and refine its own work. According to xAI, Grok 4.6 can test its outputs, identify problems, fix errors, and continue working on the task.
The model has also delivered strong results on several benchmarks. On the Artificial Analysis Intelligence Index, Grok 4.6 scored 61 points, matching GPT-5.6 Sol. It also achieved 65.9% on DeepSWE 1.1 and 69.9% on CursorBench v3.2.
The biggest improvements are reported in coding, web development, and AI-agent performance. xAI has tested the model on tasks ranging from turning an idea into a working application prototype to handling complex software-development workflows.
Grok 4.6 is currently available through Grok Build and Cursor, as well as via API. API pricing starts at $2 per 1 million input tokens and $6 per 1 million output tokens.
The key shift with Grok 4.6 is its focus on AI agents that can plan, execute, test, and improve their work over extended tasks — rather than simply answering individual prompts.
