xAI Launches Grok 3 to Rival OpenAI and Google in AI Race
Elon Musk's xAI officially releases its flagship Grok 3 model, claiming it outperforms competitors on major industry benchmarks using a massive GPU supercluster.
Elon Musk’s AI startup xAI releases its latest flagship model, Grok 3, bringing new capabilities to the Grok iOS and web applications. The company trains this new model using a massive data center in Memphis that contains approximately 200,000 GPUs. Musk claims that Grok 3 uses ten times more computing power than its predecessor, relying on an expanded training set that even includes court case filings.
The Grok 3 family includes a standard version and a faster, slightly less accurate Grok 3 mini. xAI asserts that Grok 3 outperforms OpenAI’s GPT-4o on key math and science benchmarks like AIME and GPQA. Additionally, an early version of the model scores competitively in Chatbot Arena, a popular crowdsourced platform where users vote on the best AI responses.
The release also features Grok 3 Reasoning and Grok 3 mini Reasoning, which are designed to think through problems and fact-check themselves before answering. xAI claims these specific reasoning models surpass OpenAI’s best o3-mini versions on several benchmarks, including the new AIME 2025 mathematics test. These features begin rolling out to users through the Grok app on Monday.