Here comes Grok 4: Elon Musk's $300/month artificial intelligence

In the swirling artificial intelligence market, Elon Musk has decided to raise the bar with the announcement of the latest evolution of the artificial intelligence model developed by xAI with a statement that would leave no room for reply: 'As far as academic matters are concerned, Grok 4, is above the level of a PhD in every subject, without exception'. During the live presentation, the South African billionaire added that although he may still lack common sense and has not yet discovered new laws of physics, 'it is only a matter of time'.

The model has two variants: the

standard and Grok 4 Heavy, a multi-agent configuration in which several instances work in parallel on the same problem, then comparing results to converge on a common, reasoned and robust solution. According to the researchers, the collaborative behaviour of the 'sub-models' allows Grok 4 Heavy to identify complex insights and refine them through continuous internal comparison, as would happen in a study group. xAI announced that a specialised programming model, a multimodal agent and a video generation system will also be released in the coming months.

During the live presentation, Grok 4's scores on various academic benchmarks were released. On the 'Humanity's Last Exam' test - a particularly tough benchmark that tests AI on 2,500 complex multidisciplinary questions, each of which requires knowledge and skills equal to or greater than those of a specialised researcher to solve - Grok 4 basic without 'tools' achieves a score of 25.4%, outperforming both Google's Gemini 2.5 Pro (21.6%) and OpenAI's o3 model (21%). Grok 4 Heavy, on the other hand, scores 44.4 per cent (against 26.9 per cent for Gemini 2.5 Pro). Even on the visual puzzles of the ARC-AGI-2 test, Grok 4 shows its muscles: the model achieved 16.2 per cent accuracy, doubling the performance of its nearest competitor, Claude Opus. As the xAI team points out, 'it was the only model in the last three months to break the 10% barrier'. The model's ability to maintain decision consistency was also put to the test in the VendingBench simulator, where Grok proved to be able to run a micro-business more profitably and stably than other models.

Grok 4 is not yet available in Italy. In some regions it can be accessed through an X Premium Plus subscription, but for Europe we will have to wait a little longer. The SuperGrok subscription costs EUR 35 per month, while access to Grok 4 Heavy and advanced features is provided by the EUR 349 per month SuperGrok Heavy plan, placing xAI among the most expensive providers in the industry. For comparison, ChatGPT Plus costs $20 per month, while Claude Pro stands at $20-25. From an API perspective, Grok 4 costs $3 per million tokens in input and $15 per million tokens in output, pricing that puts it firmly in the premium segment of the B2B market.