SpaceXAI released the Grok 4.7 model on September 21, designed for programming, agentic tasks, and knowledge processing. The new version features a larger base model, extended reinforcement learning, and improved handling of complex, multi-hour tasks.
In internal benchmarks, Grok 4.7 scored 46.3% in CursorBench 4.0, compared to 40.4% for the previous version. Artificial Analysis awarded the model 46 points on the Intelligence Index and 56 points on the Coding Agent Index when paired with Grok Build.
The model supports a 500,000-token context window and accepts both text and image inputs. It offers four reasoning depth levels: low, medium, high, and xhigh. For requests up to 200,000 tokens, pricing is $2 per million input tokens, $0.50 for cached input, and $6 per million output tokens.
SpaceXAI also introduced Grok 4.7 Fast, which features double the generation speed at twice the cost, available exclusively via Cursor and Grok Build. The company reported that the system's security was overhauled, achieving a 3.3% failure rate in HackerBench v0.3.
