Elon Musk’s xAI unveiled its latest artificial intelligence model, Grok 3, on Tuesday, claiming it surpasses offerings from OpenAI and China’s DeepSeek based on initial testing. The company cited Grok 3’s performance on standardized tests in math, science, and coding as evidence of its superior capabilities.
“We’re very excited to present Grok 3, which is, we think, an order of magnitude more capable than Grok 2 in a very short period of time,” Musk said during a streamed demonstration on X.
xAI also announced “Deep Search,” a “next generation search engine.” Grok 3 will be available to premium X subscribers starting Tuesday in the US, with separate subscriptions for web and app versions also offered.
Musk previously touted Grok 3’s “scary smart” reasoning abilities at the World Governments Summit in Dubai, claiming it outperformed all other models in internal tests. He noted its training on “a lot of synthetic data” and its capacity for self-reflection and logical consistency. xAI also stated that an early version of Grok 3 received higher ratings than competitors on Chatbot Arena, a crowdsourced AI model comparison website.
Musk acknowledged the model is still in beta, anticipating imperfections while promising rapid improvements. Voice assistance for Grok 3 is planned for a later release.
xAI, founded by Musk in 2023, enters a that includes OpenAI and DeepSeek. This announcement follows OpenAI’s September launch of its advanced o1 model, known for its reasoning abilities and complex task-solving. Musk, a co-founder of OpenAI in 2015, has had a strained relationship with the company’s leadership in recent years. DeepSeek recently made waves by claiming its open-source model rivaled OpenAI’s o1 in performance despite using a less resource-intensive process.
xAI’s “Colossus supercomputer,” powered by Nvidia GPUs, has been instrumental in training its models. The company revealed it recently doubled its GPU cluster for Grok 3’s training.
Related Readings:
