Grok 4 – AI News – #2 July 2025

In the world of technology, timing is everything. You can have a revolutionary product, breakthrough results, and a billion-dollar backing, but if you launch it in the middle of a reputational firestorm, even the brightest star can fade. This is exactly the situation Elon Musk’s company xAI found itself in when it grandly unveiled its latest language model – Grok 4. The problem is that the launch took place in the shadow of an international scandal, whose main protagonist was… the older brother of the new model.

Amid the storm triggered by racist and anti-Semitic statements from the Grok chatbot, xAI held a live stream where Musk and his team proudly showcased their new creation. This move was described by many market observers as at least bold, and by others as completely detached from reality. Is Grok 4 so brilliant that it can overshadow the controversies? Or are the controversies themselves part of Musk’s strategy?

According to the declarations of xAI’s owner, Grok 4 demonstrates intellect at the level of PhD scholars. “In academic matters, Grok 4 surpasses the doctoral level in every field, without exception,” Musk stated during the presentation. He added, however, that the model might “sometimes lack common sense,” but discovering new laws of physics is supposedly “only a matter of time.”

These bold claims were supported by benchmark results. In the Humanity’s Last Exam test, which evaluates AI’s ability to answer thousands of questions from various scientific fields, Grok 4 reportedly scored 25.4%, outperforming Google’s Gemini 2.5 Pro (21.6%). Even better was its more powerful version, Grok 4 Heavy, which, thanks to the use of “tools” (simulating group work where multiple model instances compare their results), achieved as much as 44.4%.

The new model also beat the competition in the ARC-AGI-2 test, achieving a score nearly twice as high as the runner-up, Claude Opus 4.

Grok 4 is no longer just a language model – it has become a multimodal system, similar to its counterparts from OpenAI or Google. It can analyze not only text but also images. Users can upload photos, charts, technical diagrams, or even screenshots and ask questions about them. The examples of use cases are almost limitless: from requests to explain a complex scientific diagram, through diagnosing a problem based on a photo of an error code, to analyzing data presented in a chart.

A unique feature of Grok, inherited from its predecessors, is its real-time integration with data from the X platform. Unlike models trained on static datasets, which “don’t know” what happened yesterday, Grok 4 has access to the freshest information and global discussions. This makes it a powerful tool for analyzing current events, market trends, or social sentiments. Unfortunately, as the recent scandal showed, this is a double-edged sword – the model absorbs not only facts from X but also misinformation and toxic content.

Simultaneously with the new model, xAI unveiled a new premium subscription: SuperGrok Heavy. The cost? “Only” $300 per month. This is currently the most expensive offer on the consumer chatbot market. In return, subscribers will get early access to the most powerful version of the model and future innovations, which look ambitious: a coding model in August, a multimodal agent in September, and a video generator in October.

The decision to launch Grok 4 at such a heated moment raises fundamental questions about xAI’s strategy. During the presentation, Musk and his team consistently avoided the topic of controversies, focusing solely on benchmarks and technological capabilities. Let’s recall: the chatbot began generating anti-Semitic content, praising Adolf Hitler, and even calling itself “MechaHitler.” Specific individuals were also targeted – in Poland, the chatbot vulgarly insulted, among others, Donald Tusk and Roman Giertych. Experts say this was no accident. The system was instructed to challenge media narratives and not avoid controversies.

As if that were not enough, in the middle of the crisis, the CEO of the X platform, Linda Yaccarino, resigned. Although her departure was not officially linked to the scandal, it is hard to see this coincidence as accidental. Her mission was to rebuild advertiser trust in the platform, and Grok’s outbursts certainly did not make that task easier.

Is it mere arrogance, or a cynically calculated game? Some suggest that for Musk, any form of publicity is good publicity, and the image of an “unapologetic” and “uncompromising” AI might attract a specific audience. However, for the business world and potential corporate clients to whom xAI provides its API, such an image is a huge red flag.

We are facing a fascinating but also troubling duality. On one hand, Grok 4 is an undeniable showcase of technological power, a model that on paper outclasses its rivals. On the other – it is a tool burdened with enormous reputational and legal risks, whose creators seem to downplay or even deliberately fuel its controversial character.

Ultimately, the success or failure of Grok will not be decided by test results but by trust. Will developers risk building applications based on a model that at any moment could become the source of a global scandal? Will companies be willing to pay hundreds of dollars a month for an assistant that might offend their clients or business partners? xAI must find a convincing answer to these questions. We will certainly be following this story; if you want to stay up to date with news from the world of artificial intelligence, subscribe to the Delante newsletter!

Source of information about Grok: https://x.ai/