xAI introduced Grok 4 and a SuperGrok Heavy subscription for $300 a month

Ilon Musk-owned xAI has unveiled its new flagship artificial intelligence model, the Grok 4, along with a $300/month SuperGrok Heavy subscription. The model was officially announced Wednesday night, after months of testing in the X ecosystem, a social network recently absorbed by xAI.

Grok is xAI’s answer to models like OpenAI’s ChatGPT and Google’s Gemini. The model can analyze images and answer questions, and as of late has been integrated more deeply into the X platform. But the popularity has also turned into attention to the AI’s controversial behavior, which has already led to incidents on the platform.

At the same time, it’s also been getting attention for its controversial behavior, which has already led to incidents on the platform.

Ilon Musk said on live TV that Grok 4 exceeds PhD level in all academic subjects without exception. He noted that the model still lacks common sense and has not yet created new technologies, but that, he said, is only a matter of time.

He said that the model has yet to make sense and has not created new technologies.

In conjunction with Grok 4, the company also introduced Grok 4 Heavy, a version with what it calls a multi-agent architecture. The model creates multiple “agents” that simultaneously solve a single problem and then compare their answers to choose the best one. This, Musk argues, is reminiscent of the work of a study group.

According to xAI, the tool-less Grok 4 scored 25.4 percent on Humanity’s Last Exam, a difficult exam consisting of thousands of questions in math, humanities, and science. By comparison, Gemini 2.5 Pro scored 21.6% and OpenAI o3 (high) scored 21%. Grok 4 Heavy with tools scored 44.4% – almost twice as much as its closest competitor.

In addition, according to Arc Prize, Grok 4 set a record on the ARC-AGI-2 test, scoring 16.2% – nearly double the previous best score among commercial AI models. This test evaluates the AI’s ability to recognize visual patterns and solve logic problems.

And it’s a test that evaluates the AI’s ability to recognize visual patterns and solve logic problems.

Along with the launch of the models, xAI introduced a new SuperGrok Heavy plan for $300 per month. It includes early access to the Grok 4 Heavy and future features. The company has promised the launch of an AI model for programming in August, a multimodal agent in September, and a video generator in October. That makes xAI’s subscription the most expensive among the major AI platforms.

So far, the company’s subscription is the most expensive among the major AI platforms.

Also, xAI plans to pursue a B2B direction and has already started providing Grok 4 via an API for developers. So far, the company’s enterprise sector has only been around for two months, but it intends to partner with major cloud platforms to expand access to the model.

Despite the impressive technical results, however, Grok’s reputation is still under pressure. A recent incident where the model’s automated account left anti-Semitic comments forced the company to step in and change system settings. xAI has since removed the section in the model’s instructions that allows “politically incorrect” remarks.

After that, xAI removed the section that allows “politically incorrect” remarks from the model’s instructions.

Whether xAI can convince businesses that Grok is reliable remains an open question.

The story xAI unveils Grok 4 and a $300-a-month SuperGrok Heavy subscription was first published on ITZine.ru.