xAI Grok 4

A long-form technical reference describes the xAI Grok 4 model family, launched July 9, 2025 and trained with reinforcement learning at pretraining scale on the Colossus 200,000-GPU supercomputer. The article enumerates six variants in the family, spanning from Grok 4.1 Fast at $0.20 per million tokens up to Grok 4 Hea The same article characterizes Grok 4 as a step change from Grok 3 in three areas — reinforcement learning applied at pretraining scale, expanded training compute via Colossus, and benchmark leadership on math and reasoning suites — and positions it as xAI's frontier offering now that xAI is described as a subsidiary o

A peer-reviewed study in Scientific Reports benchmarks 11 recent LLMs — including GPT-5, Gemini 2.5 Pro, Grok-4, and DeepSeek R1 — against 119 pediatric dentistry questions drawn from the past ten years of Turkey's Dentistry Specialization Examination (DUS). Grok-4 achieved 88.24% accuracy on the question set, statisti The study also reports response-generation times, noting that lighter models such as Gemma, LLaMA, and Mistral returned answers in under one second but at low accuracy, while reasoning-intensive modes like DeepSeek variants took longer. Grok-4's combination of high accuracy and the study's inclusion of reasoning-mode a