SpaceXAI has released Grok 4.7, securing the fourth-place spot among frontier AI laboratories in independent evaluations.
The independent benchmarking organization Artificial Analysis awarded the model an Intelligence Index score of 46, representing a 2-point increase over Grok 4.6. Higher rankings are currently held by Claude Fable 5.1, GPT-6 Astra, and Claude Opus 5.
Grok 4.7 Pushes SpaceXAI Into the Top Four AI Labs
Prior to the release, Elon Musk asserted that the model would utilize a 2.1-trillion-parameter base and surpass all existing models. He additionally noted that SpaceX company data was incorporated into the training process.
Regarding actual performance, Grok 4.7 achieved 1657 Elo on AA-Briefcase, Artificial Analysis’s benchmark designed for practical professional tasks. This reflects an 111-point improvement compared to Grok 4.6, placing the model near the performance level of Claude Opus 5.
On GDPval-AA, the model attained 1695 Elo, positioning it 90 points ahead of its predecessor. Similar upward trends were observed in programming evaluations.
When combined with its dedicated coding agent, Grok Build, Grok 4.7 received a score of 56 on the Coding Agent Index, marking a nine-point gain over Grok 4.6.
The model has surpassed GPT-5.6 Sol, leaving it behind only Anthropic’s Claude Fable 5.1, GPT-6 Astra, and Opus 5.
“Grok 4.7 scores 46 on the Artificial Analysis Intelligence Index to bring SpaceXAI into the top 4 AI labs. Coding Agent Index performance has also improved, overtaking GPT-5.6 Sol,” the post read.
Performance remained largely unchanged across other areas. The model saw a 4.5 percentage point increase on Terminal-Bench 4.0 and a 3-point rise on GDP.pdf, whereas scores declined on the AA-LCR and AutomationBench-AA tests. Grok 4.5 previously led AutomationBench in July, which supported Musk’s assertion that it performed on par with Claude Opus.
The Gains Come With a Token Bill
Pricing remains unaltered. Grop 4.7 continues to be billed at $2 per million input tokens and $6 per million output tokens. Cache hits maintain a discounted rate of $0.50, and the 500,000-token context window is carried over from Grok 4.6.
However, the model expends significantly more work per response. Artificial Analysis calculated approximately 81,000 output tokens consumed per Index task, compared to 36,000 for Grok 4.6 and 27,000 for GPT-6 Astra. Consequently, constant per-token rates still result in higher overall expenses per task.
This heavy consumption of tokens arrives for a division that has already documented a loss of $1.26 billion. Meanwhile, Musk stated on September 14 that training for Grok 4.8 is slated to conclude within a week.
He also anticipates that Grok 5 will achieve artificial general intelligence. The upcoming launch will evaluate whether SpaceXAI can advance in the standings without incurring additional compute costs to do so.
Subscribe to our YouTube channel to watch leaders and journalists provide expert insights