xAI releases Grok 4.7 after delays with 2.1 trillion parameters

xAI released Grok 4.7 on Monday afternoon and called it a notable improvement over Grok 4.6 at the same price and speed. The model is immediately available in the Grok app, Cursor, Grok Build and the xAI API, with no waitlist. It has 2.1 trillion parameters, up 40% from the 1.5 trillion in Grok 4.6, and costs $2 per million input tokens and $6 per million output tokens. xAI says Grok 4.7 spends longer working through difficult problems, double-checks its answers more often than Grok 4.6 and has the company's strongest safety guardrails so far. The model also received supplemental training data from SpaceX, including Starlink satellite telemetry, manufacturing records and engineering failure logs. xAI presents that data as a way to improve reasoning about hardware and physical systems. The release followed several apparent delays. Elon Musk revised the timeline at least five times since late July, describing the launch as four weeks away, a few weeks away, three to four weeks away, 10 days away on September 1 and a few more days away on September 11. Musk later described Grok 4.7 as a strong combination of intelligence, speed and low cost. On GDPval, which evaluates economically valuable knowledge work, Grok 4.7 scored 1695, while Claude Fable 5.1 scored 1735. On the AA-Briefcase test of multi-hour office work, Grok 4.7 scored 1657, compared with 1678 for Fable 5.1. On CursorBench 4.0, which measures coding performance against task cost and token use, Grok 4.7 was described as landing in the middle. It was more expensive per task than GPT-5.6, its successor GPT-6 Astra, and Claude Sonnet 5, while Fable 5.1 led at every price point shown. The article says this continues a pattern in which xAI's models have trailed leading Claude and OpenAI models on some benchmarks while competing through lower costs and broader availability. Grok 4.7 can serve users on X, in the standalone app and through Tesla's dashboard. Musk had said before launch that the model would be roughly on par with Anthropic's Claude Opus 5.0 rather than the newer Opus 5.1, and that its multimodal performance still needed work. He also described Grok 4.8 as a meaningful step up, Grok 4.9 as being in the Astra or Fable class and Grok 5 as a possible frontier leader, but gave no release dates for those models.