xAI has released Grok 4.7 for coding and knowledge work at prices well below many Western frontier models. Access is available through the Grok API, Cursor and Grok Build, with usage priced at $2 per million input tokens and $6 per million output tokens.
The company says the model uses a larger base, longer reinforcement learning and improved self-verification. Independent results cited by The Decoder suggest those changes have not put it at the top of the market. Grok 4.7 scored 46 on version 4.3.2 of the Artificial Analysis Intelligence Index, compared with 53 for both Claude Fable 5.1 and GPT-6.
The difference was larger on Terminal-Bench 4.0, which tests agents performing software tasks in a terminal. Grok reached 26 percent, while GPT-6 Astra scored 60 percent and Claude Fable 5.1 scored 55 percent. The lower-cost DeepSeek V4.1 Flash narrowly exceeded Grok at 27 percent.
Benchmarks measure selected tasks rather than every production workload, and xAI’s claims about training do not establish how the model will behave in a particular application. The practical trade-off is therefore price against measured capability: developers can access a comparatively inexpensive new model, but should run their own evaluations before assuming it can replace stronger systems for autonomous coding.