According to Musk, the model is designed to learn from its mistakes, adjusting its answers and improving its logical thinking
Elon Musk’s AI company, xAI, has launched Grok 3, a new flagship model that rivals OpenAI’s GPT-4o and Google’s Gemini. Grok 3 can analyze images and answer questions.
What’s new?
Grok 3 took several months to develop and the model was trained in a data center with 200,000 GPUs. Musk states that Grok 3 used 10 times more computing power than its predecessor, Grok 2, and was trained with a more extensive data set.
Grok 3 includes several models, including Grok 3 mini, which is more responsive but less accurate. xAI introduces the Grok 3 Reasoning models, which fact-check themselves for better reliability. According to xAI, Grok 3 outperforms OpenAI’s GPT-4o on benchmarks such as AIME and GPQA. The “thoughts of the reasoning models are not available to prevent distillation. Recently, DeepSeek was accused of distilling OpenAI’s models to train its own.
A new feature called DeepSearch helps users with online research by analyzing information and generating summaries. Premium+ subscribers of X are the first to get access to Grok 3, while more advanced features are behind a SuperGrok subscription.
Soon xAI will add a “voice mode,” and in a few weeks Grok 3 will become available via an enterprise API. Musk also announced that xAI will open-source Grok 2 as soon as Grok 3 is stable.