According to reports from Artificial Analysis, SpaceXAI's Grok 4.7 achieved a 46 score on the Artificial Analysis Intelligence Index, placing it fourth among frontier AI labs. The model scored 1657 Elo on AA-Briefcase, representing an 111-point jump over Grok 4.6 and bringing its performance close to Claude Opus 5. On GDPval-AA, Grok 4.7 reached 1695 Elo, which is 90 points ahead of its predecessor. The model also scored 56 on the Coding Agent Index, nine points above Grok 4.6, and passed GPT-5.6 Sol while trailing only Claude Fable 5.1, GPT-6 Astra, and Opus 5. At the current Intelligence Index version 4.3, Grok 4.7 sits one point behind GPT-5.6 Sol (47), two behind Muse Spark 1.3 (48), and four behind Claude Fable 5 (50).
As reported by Artificial Analysis, Elon Musk had stated before launch that Grok 4.7 would run on a 2.1-trillion-parameter base and exceed all current models. The model training included SpaceX company data. According to the benchmarking firm, Grok 4.7 scored 46 on the Artificial Analysis Intelligence Index, 2 points above Grok 4.6. The model features a 500,000-token context window and maintains the same pricing structure as its predecessor. Claude Fable 5.1, GPT-6 Astra, and Claude Opus 5 all rank higher in the current standings, with Claude Fable 5.1 and GPT-6 Astra leading at 53 apiece.
According to Artificial Analysis, Grok 4.7 burns 125% more tokens than its predecessor, despite maintaining the same rate card. The model costs $2 per million input tokens and $6 per million output tokens, with cache hits discounted to $0.50. Artificial Analysis measured approximately 81,000 output tokens per Intelligence Index task for Grok 4.7, compared to 36,000 for Grok 4.6 and 27,000 for GPT-6 Astra. This increased token consumption occurs on a division that has already reported a $1.26 billion loss. The model averages about 7.1 minutes per Intelligence Index task with answer output at approximately 188 tokens per second for long prompts.
Grok 4.7 demonstrates exceptional performance in agentic knowledge work tasks, with Artificial Analysis reporting that the model scores 1994 Elo for analytical quality and 1499 for presentation quality, compared with 1690 and 1519 respectively for Grok 4.6. The model significantly improves across all three components of the DeepSWE v1.1 evaluation: DeepSWE rises from 65% to 73%, Terminal-Bench 4.0 from 18% to 33%, and SWE-Atlas-QnA from 58% to 63%. These results evaluate Grok with Grok Build, its first-party coding agent, and demonstrate the model's ability to produce practical work products such as documents, spreadsheets, and slides. The model's performance on AA-Briefcase, which evaluates long-horizon agentic knowledge work, positions it just behind Claude Opus 5 and Claude Fable 5.1 at the frontier.
As reported by Artificial Analysis, Musk stated on September 14 that Grok 4.8 would finish training within a week. He also expects Grok 5 to be the model that reaches artificial general intelligence. The next release will test whether SpaceXAI can climb the rankings without burning additional compute to do so, as the company continues to balance performance improvements with cost efficiency concerns. For a company that has increasingly pitched Grok as a coding model, the 56 score on the Coding Agent Index ranks fourth among native harnesses, ahead of GPT-5.6 Sol, positioning it well for enterprises evaluating models on actual work output rather than benchmark totals.
