Elon Musk's xAI has introduced the Grok 3 series of models during a live-stream event, highlighting that the new Grok 3 model possesses 10x more compute power than its predecessor, positioning it against competitors like ChatGPT 4-o and Gemini 2 Pro. The xAI team emphasized that Grok 3 is exceptional, with even the Grok 3-Mini model performing impressively compared to rivals.
In addition to the pre-trained models, xAI launched two reasoning models: the Grok 3 reasoning model (beta) and the Grok 3 mini reasoning model. These reasoning models are designed to emulate human-like thinking, which results in longer response times but greater utility for complex queries. xAI advocates for the necessity of reasoning models, stating that merely having pre-trained models is insufficient for developing advanced AI that can think critically, evaluate solutions, and apply first principles.
xAI also noted that the Grok 3 Mini reasoning model has undergone longer training than its larger counterpart, leading to superior performance on certain benchmarks. The company claims that its reasoning models outperform OpenAI's o1, o3 Mini (High), Gemini Flash thinking, and DeepSeek R1 models in areas such as Math, Science, and coding benchmarks.
