SpaceXAI has unveiled Grok 4.7, a new model designed for programming and knowledge-based tasks. The company reports that the model utilizes a larger base architecture and extended reinforcement learning, resulting in enhanced performance on complex, long-duration tasks compared to the previous Grok 4.6 version.
In performance benchmarks, Grok 4.7 showed improvements in CursorBench 4.0, GDPval, and AA Briefcase tests. The model features a new safety architecture, achieving a 62.4% score in the LatchBio biosafety benchmark and limiting high-risk dual-use prompts to 3.3% in HackerBench v0.3.
The model is currently available via Cursor, Grok Build, and the Grok API, with support for third-party tools and cloud platforms. SpaceXAI has also granted invitation-only access to select cybersecurity partners for red-teaming and defensive research purposes.
Pricing for Grok 4.7 starts at $2 per million input tokens and $6 per million output tokens. A high-speed version is also available, offering double the output speed at twice the standard price.
