Grok 4.7: xAI’s latest model falls short in Agentic Coding.

Grok 4.7 was published by xAI on September 21, 2026. In the Terminal-Bench 4.0 benchmark, which measures a model’s ability to perform agent coding tasks (i.e., writing code and using tools autonomously), Grok 4.7 achieved a score of 26%. By comparison, Claude Fable 5.1 scored 55% and GPT-6 Astra scored 60%. This model costs $2/million input tokens and $6/million output tokens (tokens are the units of text that an AI processes; in this case, they’re the price of each unit of text processed).

A difference of 26% vs. 55% is huge when it comes to agent tasks (tasks that require autonomous action rather than simple information retrieval). DeepSeek V4.1-Flash scored 27% on the same benchmark, making it actually better than Grok 4.7. I wouldn’t recommend using this model for anything other than simple tasks, and certainly not for an agent workflow where autonomous steps need to be completed.