SpaceXAI has introduced the Grok 4.7 AI model, specifically optimized for programming tasks and scientific research. The model underwent extended reinforcement learning on complex, long-duration tasks, improving its ability to manage long contexts and verify its own output.
In the CursorBench 4.0 benchmark, the model demonstrated high price-to-performance efficiency. Grok 4.7 also features improved performance in professional fields, including legal and financial analysis, as measured by the GDPval and AA Briefcase tests.
The model includes updated security protocols. In the LatchBio test, it achieved a 62.4% success rate in handling dual-use queries. Additionally, in HackerBench v0.3, it blocked only 3.3% of legitimate requests while maintaining high resistance to malicious inputs.
Grok 4.7 is currently available via the Grok API, Cursor, Grok Build, and various cloud platforms. Pricing starts at $2 per million input tokens and $6 per million output tokens, with an optional high-speed tier available at double the cost.
