xAI introduces Grok 4 and Grok 4 Heavy as its most powerful AI models yet

xAI released Grok 4 and Grok 4 Heavy along with a new 300$ subscription plan. Grok 4 Heavy is a multi-agent system which is able to achieve a 50% score on the HLE benchmark.

· 2 min read
Grok

Elon Musk’s xAI officially released Grok 4, skipping the anticipated 3.5 update in favor of a significant leap in AI performance. The launch introduced two tiers: the base Grok 4 and a more powerful variant, Grok 4 Heavy, bundled into a new “SuperGrok Heavy” subscription priced at $300/month, targeting power users who require cutting-edge capabilities.

grok

Developed using xAI’s Colossus supercomputer in Memphis, Grok 4 features advanced reasoning, enhanced natural language precision, and a specialized coding model called Grok 4 Code, slated for release in the coming weeks. The model supports multimodal inputs—text, images, and likely soon video—and can interpret memes, offering deeper integration of cultural context than earlier versions. Benchmark results are notable: Grok 4 scored 44.4% on ARC‑AGI‑2 with tools, outperforming Gemini 2.5 Pro’s 26.9%, and nearly doubled the next-best model on visual reasoning evaluations.

grok

This version is aimed at developers, researchers, and technical enthusiasts seeking robust reasoning, scientific insight, and advanced code analysis. Elon Musk has claimed that Grok 4 could begin inventing new technologies by late 2025 and potentially achieve physics breakthroughs by 2026. Access is available through X Premium or SuperGrok subscriptions, with “Heavy” users receiving early previews of future capabilities such as video generation.

grok

xAI has addressed prior criticism surrounding Grok’s antisemitic and extremist outputs by removing problematic system prompts and strengthening moderation measures ahead of Grok 4’s release.

Founded in late 2023, xAI built a complete AI infrastructure in under two years. The Colossus supercomputer—active since December 2024—now supports all model training, though concerns remain regarding its energy consumption and reliance on methane-powered systems in Memphis. With Grok 4, xAI positions itself to directly challenge OpenAI’s ChatGPT, Google DeepMind’s Gemini, and Anthropic’s Claude, focusing on open-ended reasoning and meme fluency.

Overall, Grok 4 represents a major advancement: domain-specific reasoning, multimodal intelligence, coding expertise, and industry-leading benchmark results. As it expands across premium tiers, xAI is working to balance innovation with responsibility, continuing its push toward Musk’s goal of a self-inventing Grok AI.