File photo.
Washington: The artificial intelligence company xAI, founded by American billionaire Elon Musk, announced the release of its latest AI model, Grok 4.
This new model represents a significant step forward in the company’s quest to develop advanced reasoning systems. Designed to tackle complex logic tasks, mathematical equations, and code generation, Grok 4 marks a leap in capability over its predecessors.
The company confirmed that Grok 4 was trained using a supercomputer named Colossus, first activated in September with 100,000 graphics processing units (GPUs).
By May, the system scaled up to over 200,000 GPUs, and future plans indicate that xAI aims to reach an ambitious benchmark of one million units.
This massive computational power has enabled Grok 4 to process inputs of up to 256,000 tokens, allowing it to analyze both texts and images with impressive precision.
Beyond its technical specifications, Grok 4 has demonstrated superiority in various benchmark tests.
In a notoriously difficult assessment known as Humanity’s Last Exam, the model correctly answered more than 44 percent of the questions without using external search engines. For comparison, OpenAI’s Deep Research tool managed only 26.6 percent under the same conditions.
Elon Musk praised the model’s performance on his platform X, noting that Grok 4 rarely falters in physics or math unless the questions are deliberately misleading. He added that the model can autonomously identify and correct flaws in both questions and answers.
In parallel, xAI launched a specialized version of the model named Grok 4 Heavy, which utilizes intelligent agents working collaboratively to interpret queries, refine responses, and select the best answer.
Looking ahead, xAI intends to release a programming-specific variant in August, followed by a multimodal model in September, and eventually a video-generation version.
The company also reported ongoing efforts to integrate Grok 4 with engineering and scientific tools, such as computational fluid dynamics systems used in the development of Tesla vehicles. Distribution through public cloud platforms is part of the broader deployment strategy.
