Working with Grok? Elon Musk's crew has some upgrades for you

SpaceXAI has launched Grok 4.7, its upgraded flagship model, offering a leap in programming, engineering, and legal task performance. The model operates twice as fast and at half the cost of its competitors, surpassing GPT-6 in office work metrics. The launch includes an advanced safety suite and availability in the Cursor tool.

SpaceXAI, the artificial intelligence company owned by Elon Musk, launched yesterday (Monday, September 21, 2026) Grok 4.7 – its most powerful flagship model to date. The model focuses on programming tasks, engineering development, and complex knowledge work, setting a new benchmark for cost-effectiveness. In the CursorBench 4.0 test for long-term programming, the model achieved 46.3% at a low cost per task; in the DeepSWE v1.1 test for complex software engineering, it reached 71%. In the Harvey legal agent test, Grok scored 19.6%, compared to only 2.5% for OpenAI's GPT-5.6 Sol. In electrical engineering (EEBench), it jumped to 64%, and in the GDPval index for office tasks, it set a record of 1,695 points – above GPT-6 Astra. The model includes a rigorous self-verification mechanism to prevent errors, natural understanding of the Grok Bot interface, and handling of extremely long contexts. In security, a new protection layer approved only 3.3% of malicious requests in the HackerBench test, and it led the LatchBio biological safety index with 62.4%. The price remains $2 per million input tokens and $6 per million output tokens – identical to the previous version and tens of percent cheaper than competitors. The model is available from today in Cursor, Grok Build, API, command line, and cloud computing.