xAI launched its next-generation model, Grok 4.7, on Monday, integrating it simultaneously into the Grok app, Cursor, Grok Build, and the xAI API. The company stated that this is currently the most powerful version, offering improvements over Grok 4.6 without changes in pricing or response speed.
This release has been delayed multiple times from its original schedule. Since late July, Musk has adjusted the launch date at least five times, with the final launch occurring on September 21.
Parameter scale increases to 2.1 trillion
Grok 4.7 has a parameter scale of 2.1 trillion, representing approximately a 40% increase from Grok 4.6’s 1.5 trillion. Pricing is set at $2 per million tokens for input and $6 per million tokens for output.
xAI states that the new model will spend more time reasoning on complex problems, review its answers more frequently, and includes enhanced security settings.
Join SpaceX engineering data training
A new change in Grok 4.7 compared to previous versions is the addition of supplementary training data from SpaceX, including Starlink satellite telemetry, manufacturing records, and engineering fault logs.
xAI aims to enhance the model's understanding of hardware systems and physical engineering problems, rather than relying solely on internet text for training. This also positions Grok 4.7 more toward engineering and real-world system reasoning.
Multiple tests have still not ranked first.
Based on the data disclosed by xAI, Grok 4.7 ranked highly in multiple tests but did not reach first place. In the GDPval test, it scored 1695, below Claude Fable 5.1’s 1735; in the AA-Briefcase test, Grok 4.7 scored 1657, also below the latter’s 1678.
On the electrical engineering benchmark EEBench, Grok 4.7 lags behind GPT-6 Astra. The report also notes that on CursorBench 4.0, which evaluates real-world programming tasks, Grok 4.7 performs at an intermediate level and has still not caught up to Anthropic’s Fable 5.1.
xAI continues to bet on low-cost coverage
From Grok 4.5 to Grok 4.7, xAI’s strategy has not changed significantly. Rather than directly competing for the top spot, the company emphasizes cost, speed, and distribution channels, aiming to expand adoption through lower prices and broader reach.
Before the release, Musk stated that Grok 4.7's overall performance is roughly comparable to Claude Opus 5.0, and that its multimodal capabilities still require further improvement. He also mentioned that subsequent versions, including Grok 4.8, Grok 4.9, and Grok 5, are planned, but no specific release dates have been provided.