Elon Musk’s, xAI, on Wednesday night launched Grok 4, its most advanced language model to date, amid heightened scrutiny following a scandal involving antisemitic content generated by its earlier version. Musk used a livestream demo to showcase Grok 4’s capabilities claiming it now leads the global AI race.
Speaking alongside xAI staff, Musk boldly declared, “This is the smartest AI in the world.” He claimed Grok 4 outperformed OpenAI’s o3 (high) and Google’s Gemini 2.5 Pro in multiple AI benchmarks, including the widely cited “Humanity’s Last Exam”, a 2,500-question assessment covering maths, humanities, and science, where Gemini currently tops the leaderboard. Musk said Grok 4 would ace the SAT “every time” and even described it as “smarter than almost all graduate students, in all disciplines, simultaneously.”
Multiple Grok 4 Models
Two versions of the model were unveiled: Grok 4 and a more powerful Grok 4 Heavy, available via monthly subscriptions at $30 and $300, respectively. Musk went further, claiming Grok would not only outperform humans in academics, but could begin discovering new technologies and even new physics as early as next year. Musk however acknowledged its occasional lack of “common sense,” likening AI to “a super genius child” that needs proper values instilled for safety and alignment.
No Mention of Grok 3 Controversy
During the hour long presentation, there was no direct reference to the controversy that erupted just a day earlier, when xAI’s chatbot posted a series of antisemitic messages on X (formerly Twitter), including praise for Adolf Hitler and claims that “MechaHitler mode” was its default setting.
One particularly disturbing message said Musk “built me this way from the start.” Though the offensive posts were quickly removed, and xAI released a statement saying it had updated moderation protocols, the incident raised urgent questions about the risks of deploying powerful AI tools at scale.
In the aftermath, xAI confirmed it had revised Grok’s system prompts removing recent instructions that encouraged responses not to “shy away from politically incorrect claims.”
During the livestream, Musk touched briefly on AI safety, stating that the most important safeguard was ensuring the model is “maximally truth-seeking.” He stressed the need for AI systems to be “truthful, honorable” and aligned with the values one would instill in a highly intelligent child.
