Grok 4.7 was officially released. SpaceXAI positioned it as the latest flagship model for programming and knowledge work, and provided a straightforward upgrade logic: compared to Grok 4.6, while keeping the standard API price and service speed unchanged, the capabilities for long tasks, agent programming, and professional knowledge work have all improved.
Upgrades at the same price have widened the performance gap
The standard API pricing for Grok 4.7 remains $2 per million input tokens and $6 per output token, the same as Grok 4.6. The official company also offers a fast version with double speed, corresponding to double the price.
Performance gains are concentrated in complex tasks. CursorBench 4.0 increased from 40.4% in Grok 4.6 to 46.3%; Terminal-Bench 4.0 increased from 20.3% to 38.0%, nearly doubling. AA Briefcase scores also rose from 1546 to 1657.
Grok 4.7 is not just about changing model parameters
SpaceXAI stated that Grok 4.7 uses a larger foundational model than Grok 4.6 and undergoes longer reinforcement learning training. Training tasks focus more on problems that take hours to complete, while strengthening the model's self-checking, long-context management, and continuous execution capabilities.
This shifted Grok 4.7's goal from "answering smarter questions" to "completing complex tasks." The official focus is especially on professional work scenarios such as documents, presentations, law, electrical engineering, and long-term terminal operations, with native Grok Bot adaptation included in training.
AI competition began to calculate task costs
The key to Grok 4.7 is not to top a single benchmark, but to improve effective task completion rates while maintaining the Grok 4.6 price system. For Coding Agents, token unit price, inference speed, and the probability of completing a task are becoming more important than simply model scores.
Data from third-party Artificial Analysis shows Grok 4.7 xhigh has an IQ of 46 and an output speed of about 38.8 tokens/s. It does not lead across all comprehensive metrics, but has established strong performance density within a similar price range.
Grok 4.7 has entered development environments such as Grok API, Cursor, Grok Build, and GitHub Copilot. The large model race is shifting from "who is smarter" to "who can complete longer, more realistic work at lower cost," which may be the most noteworthy change for this generation of Grok.
