Grok 4 is xAI's frontier model and the strongest option on the Grok provider. It made headlines at launch for its reasoning benchmarks and brings that horsepower to Magnus with the family's trademark directness intact.
Launched July 9, 2025 by xAI. Grok 4 was trained with heavy reinforcement learning on reasoning, and its Heavy configuration was the first model to break 50% on Humanity's Last Exam - a 2,500-question, PhD-level benchmark spanning maths, physics, chemistry, linguistics, and engineering.
On the All-inclusive plan a Grok 4 reply costs 3 credits.
Hard reasoning with attitude: research questions, devil's-advocate takes, and debate rounds against a Claude agent. Its direct, slightly irreverent style makes it a favourite for a challenger persona on the agent bar.