SpaceXAI Presents Language Model Grok 4.7

SpaceXAI has released its flagship artificial intelligence model Grok 4.7, focused on programming and solving complex professional tasks. The developers of the company reported this.

The new model is based on a larger model compared to Grok 4.6. During its preparation, the developers increased the duration of training with reinforcement and focused on tasks that require continuous multi-hour calculations. The model received a context window of 500 thousand tokens, learned to more detailed verify the results obtained and received built-in support for the Grok Bot environment.

According to the company, in the CursorBench 4.0 programming test, the model scored 46.3%, and in the DeepSWE v1.1 benchmark, it showed a result of 71%. In the GDPval professional tasks test, Grok 4.7 scored 1695 points. Also, the system of protective mechanisms has been redesigned: in the HackerBench v0.3 cyber threat test, the share of missed dangerous requests was 3.3%, and in the LatchBio biosecurity test, the indicator reached 62.4%.

The basic cost of using the API remained at the level of the previous version - $2 for 1 million input tokens and $6 for 1 million output tokens. In addition, a Fast version is offered with double the generation speed at a double rate. The model is available through the API, as well as in Cursor and the Grok Build environment.