xAI releases Grok 4 with scaled reinforcement learning and native tool use

xAI has released Grok 4, a new frontier model trained using reinforcement learning on its 200,000 GPU Colossus cluster to refine reasoning abilities at pretraining scale. The update introduces native tool use, including real-time web search and code interpretation, alongside a new SuperGrok Heavy tier featuring the more powerful Grok 4 Heavy variant.

• Grok 4 sets a state-of-the-art for closed models on ARC-AGI V2 with 15.9% and dominates the agentic Vending-Bench with $4694.15 net worth.
• Grok 4 Heavy leads USAMO'25 with 61.9% and is the first model to score 50.7% on Humanity's Last Exam (text-only subset).
• The model features a 256,000 context window, multimodal understanding, and an upgraded Voice Mode with real-time visual analysis capabilities.

The release aims to provide developers and subscribers with enhanced reliability and complex reasoning through scaled reinforcement learning and native tool use.