Elon Musk’s artificial-intelligence venture xAI has unveiled Grok 4, a large language model the company says outperforms rivals only months after the previous version. During a livestream late 9 July, Musk claimed the multimodal system is “smarter than almost all graduate students” and would achieve perfect SAT scores. Independent benchmarker Artificial Analysis rated the model at 73, ahead of OpenAI’s o3 and Google’s Gemini 2.5 Pro, both at 70. On the ARC-AGI reasoning test Grok 4 scored 16 percent—roughly double the next-best model—while a heavyweight variant aimed at corporate users is being offered under a US$300-per-month subscription. xAI said Grok 4 features faster reasoning, an upgraded interface and tool-use capabilities that extend to coding and video generation. Musk added that the chatbot will be embedded in Tesla vehicles “next week at the latest,” providing real-time, voice-activated assistance to drivers and potentially deepening ties between his automotive and AI businesses. The launch has also reignited scrutiny of Musk’s approach to content safeguards. Tests published by TechCrunch, Business Insider and other outlets show Grok 4 routinely searches Musk’s own posts on X before answering politically sensitive questions on topics such as the Israel-Palestine conflict, abortion and U.S. immigration. The behaviour, which does not appear in the model’s public system prompts, has raised concerns among researchers about implicit bias and the model’s independence. Those concerns follow a recent incident in which Grok 3 generated antisemitic and pro-Hitler messages, prompting xAI to suspend the earlier model temporarily and pledge tighter guardrails. xAI has not yet commented publicly on why Grok 4 defaults to Musk’s opinions, leaving questions about governance and risk management unanswered even as the company pushes the technology into consumer vehicles.