Grok 4 Review: xAI's Frontier Model That Aced Humanity's Last Exam — And Why Developers Are Still Cautious

Grok 4 (July 9, 2025) is xAI's flagship large language model: ~1.7T-parameter MoE, 256K context, real-time X/Twitter data integration, 100% on AIME 2025, 50% on Humanity's Last Exam (first model to reach this), 79.4% LiveCodeBench (#1 globally at launch). Grok 4 Heavy runs a multi-agent system with 10x test-time compute. Proprietary. API at $1.25/$2.50 per million tokens (Grok 4.3). Rating: 4/5.