SpaceXAI priced its newest model closer to a Chinese lab than a Western one. Grok 4.7 costs $2 per million input tokens and $6 per million output tokens, and on the independent Artificial Analysis Intelligence Index it manages a score of 46, behind both Claude Fable 5.1 and GPT-6.
The company calls it the strongest release yet for coding and knowledge work. Training ran on a bigger base model, for longer, with a self-verification loop wired into the pipeline.
Agentic coding is where the deficit shows. Terminal-Bench 4.0 separates the leaders sharply: GPT-6 Astra scores 60 percent and Claude Fable 5.1 lands 55. Grok 4.7 reaches 26. Even DeepSeek’s cheaper V4.1 Flash takes a higher seat, at 27.
Developers can reach it through the Grok API, Cursor and Grok Build. For SpaceXAI, cadence is the strategy: frequent point releases pushed at once into agents, speech and satellite compute rather than chat alone.
