Elon Musk and the executives of xAI have launched Grok 4, the new version of the American firm’s language model, which has been presented as “the world’s smartest AI,” as it is capable of surpassing PhD levels.
The American magnate unveiled his flagship artificial intelligence model this Thursday, along with a new subscription plan for $300 a month (about €255), which includes its most advanced version, Grok 4 Heavy.
Musk began the presentation, broadcast on xAI’s official channels, by asserting that Grok 4 is “the world’s smartest AI,” as it is smarter than “almost all graduate students in all disciplines.”
In fact, the owner of X and Tesla has specified that it even surpasses PhD levels in every subject, “without exception.” However, Musk clarified that Grok 4 is not yet capable of inventing new technologies or making breakthroughs by discovering theories in the field of physics.
“But this is just a matter of timing. I think it could discover new technologies as soon as this year, and I would be surprised if it didn’t happen next year. So I expect Grok to discover new technologies no later than next year, and maybe by the end of this year,” Musk stated.
GROK 4 AND GROK 4 HEAVY
xAI has launched two models: Grok 4 and Grok 4 Heavy. The latter is a multi-agent version of the company that offers greater performance, as it has several agents working on a solution simultaneously. It is also capable of comparing their work “like a study group” to find the best answer.
The xAI leaders who accompanied Musk showed various tests they have performed with Grok 4 to demonstrate the model’s performance. One of them is the well-known Last Test of Humanity, one of the most rigorous benchmarks for measuring the capabilities of advanced language models, consisting of more than 2,500 questions on topics such as math, science, and linguistics.
Specifically, the company has stated that Grok 4 Heavy achieved a score of 44.4 percent when equipped with additional tools (external capabilities of the model that developers can add), while Google Gemini 2.5 Pro obtained a score of 26.9 percent with the same configuration.
Musk has noted that Grok’s biggest weakness currently is that it is “partially blind,” given that its comprehension and image generation “need to be much better.” The magnate has assured that work is being done on this and that the model is being trained to improve this aspect.
The launch of Grok 4 comes at a difficult time for the company, which owns xAI and Elon Musk. X CEO Linda Yaccarino resigned on Wednesday following numerous controversial comments posted by the chatbot Grok on the social network.
In Grok’s latest unusual post on X, he called himself “MechaHitler,” a reference to a 1990s video game. This comment follows those made last weekend, in which he made anti-Semitic remarks.
