The company AI de Elon Musk, XAI, on Wednesday launched on Wednesday its latest model of the badge, Grok 4, and presented a new AI subscription plan of $ 300 per month, Supergrok Heavy.
Grok is XAI’s response to models such as Openai and Google’s Chatgpt, and you can analyze images and answer the questions. In recent months, Grok has been more deeply integrated into the social network of Musk, X, which was recently acquired by XAI. However, that has also put the front and center of bad behavior of Grok for millions of users.
The expectations are high for Grok 4. The last model of Xai will be stacked against the next model of Opense, GPT-5, which is expected to be launched at the end of this summer.
«Regarding academic questions, Grok 4 is better than the doctoral level on each topic, without exceptions,» Elon Musk said during a live broadcast on Wednesday night. «Sometimes, it can lack common sense, and has not yet invented new technologies or has discovered a new physics, but that is just a matter of time.»
The launch of Grok 4 occurs in the middle of a tumultuous week for Elon Musk companies. Earlier on Wednesday, Linda Yaccarino resigned from her role as CEO of X after approximately two years with the company. X has not yet announced to his successor.
Yaccarino’s departure occurs a few days after Grok’s automated account responded to users with anti -Semitic comments that criticize Hollywood’s «Jewish executives» and praised Hitler. XAI had to briefly limit the Grok account and eliminate offensive publications. In response to the incident, XAI seemed to have eliminated a recently aggregate section of the Grok public system message, a list of instructions for the AI chatbot to follow, which told him not to avoid the «politically incorrect» statements.
The leaders of Musk and XAI greatly avoided discussing the incident, instead of focusing on the performance and capabilities of Grok 4.
XAI launched two models on Wednesday: Grok 4 and Grok 4 Heavy, the latter being the «version of multiple agents» of the company that offers a higher performance. Musk said that Grok 4 weighs multiple agents in genius to work on a problem simultaneously, and then all compare their work «as a study group» to find the best answer.
XAI states that Grok 4 shows the performance of the border level at several reference points, including the last examination of humanity, a challenging test that measures the AI ability to answer thousands of crowdsourcing questions about subjects such as mathematics, humanities and natural sciences. According to Xai, Grok 4 obtained 25.4%in the last examination of humanity without «tools», surpassing Gemini 2.5 Pro from Google, which obtained 21.6%, and O3 (High) of OpenAi, which obtained 21%.
XAI states that Grok 4 Heavy, with «tools», could achieve a 44.4%score, surpassing Gemini 2.5 Pro with tools, which obtained 26.9%.
The non -profit arc award says that Grok achieves a new latest generation score In its ARC-AGI-2 test, another difficult reference point that consists of puzzle problems where an AI has to identify visual patterns, writing down 16.2%. That is almost double the score of the next best commercial model, Claude Opus 4.
Together with Grok 4 and Grok 4 Heavy, XAI launched its most expensive IA subscription plan so far, a subscription of $ 300 per month called Supergrok Heavy. The subscribers of the plan will obtain an early preview for Grok 4 Heavy, as well as early access to new functions. The plan is similar to the ultra premium levels offered by Operai, Google and Anthrope, but Xai now offers the most expensive subscription among the main suppliers of AI.
Supergrok Heavy subscribers can get early access to some new products that XAI plans to launch in the coming months. The company said Wednesday that an AI coding model will arrive in August, a multimodal agent in September and a video generation model in October.
XAI is launching Grok 4 through its API in an effort to get developers to create applications with the model. The company indicates that the XAI business sector is only two months old, however, it plans to work with hyperscalers to make Grok available through its cloud platforms.
Despite Grok’s border level performance at reference points, it can be difficult for XAI to overcome its recent setbacks, since it tries to present Grok to companies as a true contender for Chatgpt, Claude and Gemini. It remains to be seen if companies are ready to adopt Grok, defects and everything.
