Grok 4 AI Launched: Elon Musk Claims It Surpasses Graduate-Level Reasoning The company also introduced Grok 4 Heavy, a version designed to handle more complex tasks by using a team-based system of AI agents
You're reading BIZ Experiences India, an international franchise of BIZ Experiences Media.
xAI, the artificial intelligence firm founded by Elon Musk, has released its latest model, Grok 4, in a move that signals the company's increasing pace of development in the competitive AI race. Musk claimed the new model outperforms most graduate students across academic disciplines and is capable of answering complex questions at a PhD-level.
According to him, Grok 4 demonstrates significantly improved performance compared to previous versions, though he acknowledged the model can still make errors and occasionally lack basic reasoning. He suggested it may not be far from reaching the point of technological or scientific discovery, though this has yet to be realised.
The company also introduced Grok 4 Heavy, a version designed to handle more complex tasks by using a team-based system of AI agents. These agents work independently on the same task and then compare outputs to determine the most accurate response, a method intended to improve both efficiency and reliability in problem-solving.
xAI reports that Grok 4 has achieved higher scores than some competing models in a number of standardised AI tests. On Humanity's Last Exam—a test covering maths, science, and humanities—Grok 4 scored 25.4 per cent without tool assistance. In comparison, OpenAI's o3 model scored 21 per cent, and Google's Gemini 2.5 Pro scored 21.6 per cent. Grok 4 Heavy, when tested with tools, scored 44.4 per cent, ahead of Gemini's 26.9 per cent.
Another evaluation, ARC-AGI-2, which measures how well AI can handle visual pattern problems, showed Grok scoring 16.2 per cent, which was higher than other leading models such as Claude Opus 4.
Grok 4 is capable of processing text, images, and voice inputs, and is integrated into the social media platform X (formerly Twitter), owned by Musk. Earlier versions of Grok were met with mixed responses from users, particularly regarding accuracy and usability.
As part of the rollout, xAI also announced a premium subscription tier — SuperGrok Heavy — aimed at users with advanced computational needs. The subscription is priced at USD 300 per month.
