Elon Musk’s xAI to Release Grok 4 and Grok 4 Code After July 4 With State-of-the-Art Benchmarks Including 45% on Humanities Last Exam

Elon Musk and his xAI team are preparing to release Grok 4, the latest iteration of their AI language model, shortly after July 4, 2025. Grok 4 is designed to rewrite the entire corpus of human knowledge by correcting errors, filling gaps, and retraining on cleaner data, addressing issues of bias and flawed information in previous AI models. The release will include at least two models: Grok 4, an all-rounder excelling in language, math, and reasoning, and Grok 4 Code, a specialized model focused on software engineering tasks. Leaked benchmark results suggest Grok 4 achieves state-of-the-art performance, with scores such as 35% on the Humanities Last Exam (HLE) rising to 45% when reasoning is applied, 87-88% on the GPQA benchmark, and 72-75% on the SWE Bench for Grok 4 Code.

The HLE exam is a demanding test covering 2,500 expert-written questions across more than 100 subjects, including math, physics, and computer science. Musk has emphasized the importance of accuracy and advanced reasoning in this upgrade, which he describes as the most intelligent AI model to date. Despite the anticipation and leaked data, the official release date remains shortly after July 4, with Musk continuing to work closely with the xAI team to finalize the model.