SpaceXAI has launched Grok 4.7, a new base model that maintains the same pricing structure as its predecessor, Grok 4.6. The model is priced at $2 per million input tokens and $6 per million output tokens. It is currently available via the xAI API, Cursor, Grok Build, OpenRouter, Vercel, and Cloudflare.
The model demonstrates performance improvements across several benchmarks compared to Grok 4.6. Notably, its score on Terminal-Bench 4.0 increased from 20.3% to 38.0%, and it achieved 64.0% on EEBench. In professional knowledge work testing, Grok 4.7 reached 1,695 Elo on the GDPval benchmark, up from 1,605 for the previous version.
Grok 4.7 introduces a new safeguard stack, which the company describes as its most resistant to jailbreaks and refusals to date. It achieved a 62.4% score on LatchBio’s biosafety benchmark. Additionally, on the internal HackerBench v0.3, the model allowed 3.3% of risky dual-use prompts to pass through.
A specialized version, Grok 4.7 Fast, offers double the output speed but at twice the cost. This version is restricted to Cursor and Grok Build and is excluded from the free tier. Furthermore, users can access a US regional endpoint for inference at a 10% price premium.
