xAI's Grok 4 Achieves 15.9% on ARC-AGI-2, Claims Top AI Benchmark Spot

xAI, Elon Musk's artificial intelligence company, officially launched Grok 4 on July 9, 2025, positioning it as the "world's most powerful AI model." The release introduced a new "SuperGrok Heavy" tier at $300 per month, alongside the standard $30 per month offering. The model's advanced capabilities have already garnered significant user reactions, with prominent user Mario Nawfal tweeting, "> 'Grok 4 went from the "wow" factor to now leaving us speechless.'"

Grok 4 boasts native tool use, real-time search integration, and enhanced logical consistency through first-principles reasoning. xAI claims the model achieves impressive results on various benchmarks, including a 15.9% score on ARC-AGI-2 and strong performance on GPQA, often outperforming competitors like OpenAI’s GPT-4o and Google’s Gemini 2.5 Pro. The model was reportedly trained using xAI's Colossus supercomputer, leveraging 200,000 GPUs, representing a 10x increase in compute from its predecessor, Grok 3.

The introduction of Grok 4 and its premium "Heavy" tier signals xAI's aggressive strategy in the competitive AI landscape, targeting both general users and enterprise clients. The standard Grok 4 is accessible at $30 per month, while the "SuperGrok Heavy" subscription, priced at $300 per month, provides access to the more powerful Grok 4 Heavy and higher rate limits. An API is also available, facilitating integration into various applications and developer workflows.

Grok 4's release follows previous versions that faced scrutiny over problematic content generation and controversial responses. xAI emphasizes its mission to create accurate AI systems and has reportedly implemented measures to enhance transparency and reliability. Future plans for Grok 4 include expanding multimodal capabilities with vision and image generation, and specialized models like Grok 4 Code for software development, further solidifying xAI's ambition to push the boundaries of artificial intelligence.