KOSUNI CRYPTO
Korea's Crypto Pulse, in English
$BTC $ETH $XRP $SOL $DOGE
News
Neutral

Elon Musk Claims Grok 4.7 Surpasses All Models—Does It?

Published September 22, 2026 11:59 AM · 5 views
Elon Musk Claims Grok 4.7 Surpasses All Models—Does It?

SpaceXAI has released Grok 4.7. Independent benchmarking by Artificial Analysis ranked the model as fourth among top AI labs.

The model received an intelligence index score of 46 from Artificial Analysis, two points higher than Grok 4.6. Claude 5.1, GPT-6 Astral, and Claude Opus 5 scored higher.

Elon Musk claimed ahead of the release that the model would surpass all existing models, running on a 2.1 trillion-parameter foundation with SpaceX company data included in training. He stated it would "outperform all existing models."

In practical benchmarks, Grok 4.7 scored 1,657 Elo in Artificial Analysis's AA-Briefcase specialized benchmark for professional tasks, up 111 points from Grok 4.6 and nearing Claude Opus 5 levels. It achieved 1,695 Elo on the GDPval-AA scale, a 90-point increase over its predecessor.

With its coding agent Grok Build, Grok 4.7 scored 56 in the coding agent index, up nine points from Grok 4.6. The model surpassed GPT-5.6 Sol but trails behind Claude 5.1, GPT-6 Astral, and Opus 5.

Other metrics showed mixed results: a 4.5-point increase in Terminal-Bench 4.0 and a three-point rise on GDP.pdf, but declines in AA-LCR and Automation Bench-AA. Grok 4.5 had previously beaten Automation Bench in July, supporting Musk's claim that it was "equal to Claude Opus."

The pricing remains unchanged: $2 per million input tokens, $6 per million output tokens, with a $0.50 discount for cached results and a 500k context window retained from Grok 4.6.

However, the model uses significantly more tokens per task—approximately 81,000 output tokens on average compared to 36,000 for Grok 4.6 and 27,000 for GPT-6 Astral—meaning higher costs per task despite identical pricing.

This token usage has impacted a department already reporting $1.26 billion in losses. Musk also announced that Grok 4.8's training would complete within a week on September 14 and that Grok 5 is expected to reach artificial general intelligence (AGI).

Korean Source

This article is an English localization of a Korean-language crypto news report. Original headline: 일론 머스크, 그록 4.7 ‘모든 모델 능가’…실제로 그랬나?