xAI, an AI company of businessman Elon Musk, has launched the Grok 3, a family of AI large language models that appears to outperform other AIs as per initial standardized benchmarks.
The Grok 3 models were trained using xAI’s Colossus supercomputer cluster that utilizes 100,000 NVIDIA Hopper Tensor Core GPUs. The company has introduced standard and mini non-reasoning models called the Grok 3 beta and Grok 3 mini beta, as well as a pair of reasoning models called the Grok 3 beta (Think) and Grok 3 mini beta (Think).
As per the reports, the non-reasoning model generally outperformed OpenAI GPT-4o and DeepSeek-V3. Grok has a one million token context window allowing it to use substantial amounts of text so it can formulate correct answers from different sources.
Meanwhile, the reasoning models are thinking through complex prompts step-by-step so users can see through their thought processes. These models were made to handle expert problems by solving smaller ones and then combining results for a real answer.
Selecting the DeepSearch engine will task Grok 3 to do a broad and deep search across the internet and use code interpreters. It will then summarize the findings to create a comprehensible report.
xAI will continue to develop its Grok 3 in the coming months using 200,000 GPU supercomputer clusters to improve performance. The new AI is now available on Grok.com and X users. There’s a paid version and a free one with limited capabilities.