Elon Musk’s company, xAI, has officially introduced its new flagship AI model — Grok 3. Released on February 17, 2025, it brings significant improvements over its predecessor Grok 2 and is aimed at competing with leading AI models such as OpenAI’s GPT-4o and Google’s Gemini. Grok 3 is integrated into the Grok app for iOS and web.
Grok 3 is a family of models, including Grok 3 mini, which prioritizes speed over accuracy, as well as specialized models for handling complex tasks in mathematics, science, and programming, such as Grok 3 Reasoning and Grok 3 mini Reasoning. These models perform thorough self-checks before providing answers, enabling users to access enhanced capabilities through modes that are especially suited for complex queries.
The development of Grok 3 relied on significant computational resources, including a data center in Memphis equipped with around two hundred thousand GPUs. According to Musk, the model was trained using ten times more computing power than its predecessor, on an expanded dataset that includes legal documents.
xAI claims that Grok 3 outperforms competitors on tests such as AIME (focused on mathematical questions) and GPQA (testing PhD-level knowledge in physics, biology, and chemistry). In addition, the new DeepSearch feature expands Grok’s capabilities by scanning the internet and the X social network to provide users with concise answers to their queries.
