Grok 4 Could Be a Turning Point in Multimodal AI

xAI, the artificial intelligence company founded by Elon Musk, is moving swiftly in the AI arms race with the release of Grok 4, its latest model. According to Musk, Grok 4 surpasses most graduate students in academic performance and can answer complex, PhD-level questions. The launch signals the company’s growing ambition to compete with top players like OpenAI and Google in the development of high-performing AI systems.

Grok 4 brings notable improvements over its earlier versions, particularly in processing accuracy and response depth. While Musk acknowledged the model still makes mistakes and sometimes struggles with basic reasoning, he hinted that it could be nearing the ability to make genuine scientific or technological discoveries, a milestone that remains to be proven. The new model is also multimodal, capable of handling text, images, and voice input, and is already integrated into X.

To expand its capabilities even further, xAI introduced Grok 4 Heavy, a specialized version built for more demanding tasks. This variant employs a team-based system of AI agents that tackle the same problem independently before comparing results to select the most accurate answer. The approach is designed to improve both efficiency and reliability, particularly in high-stakes or complex problem-solving scenarios.

Grok 4’s performance metrics reflect its rapid progress. On Humanity’s Last Exam, a comprehensive test of math, science, and humanities knowledge, Grok 4 scored 25.4% without tool assistance, outperforming OpenAI’s o3 (21%) and Google’s Gemini 2.5 Pro (21.6%). For users with advanced computational needs, xAI also launched SuperGrok Heavy, a premium subscription tier priced at $300 per month.