Elon Musk’s xAI has just unveiled its latest artificial intelligence model, Grok 3, and early testing, which included standardized tests on math, science, and coding, shows it outperforms offerings from OpenAI and China's DeepSeek. On top of that, compared to their previous model, Elon claims that Grok 3 is a “magnitude more capable than Grok 2 in a very short period of time.” During the presentation, the xAI team demonstrated Grok 3's capabilities by solving live physics problems, creating a game from scratch, and much more.

Apart from unveiling Grok 3, there were a ton of other showcases, one of which was a new product called “Deep Search,” which would act as a “next-generation search engine”. Grok 3 has now been released for Premium+ X subscribers and will also be accessible through a separate subscription for the model's web and app versions.

Grok 3 leads as the smartest AI model in benchmarks

Early reviewers, like Lex Fridman from the Lex Fridman Podcast, are impressed with Grok 3’s exceptional performance, as the AI model outperforms every other model based on benchmarks. For instance, in the AIME’ 24 test, Grok 3 leads with a score of 52, dominating OpenAI’s GPT-4, which had only 9 points. In the GPQA test, the model once again topped the charts with a whopping score of 75, and in LCB, Grok 3 scored the highest with 57. What’s even more interesting is that in all three of these tests, the runner-up was also xAI's Grok 3 Mini.

Part of the reason why Grok 3 is leading with such exceptional margins over models like DeepSeek is that it's powered by a supercomputer with over 100,000 Nvidia GPUs, called Colossus. But the company has now doubled the size of its GPU cluster. According to xAI, Grok 3 was trained on Colossus with its phase 2 extension, ramping up to 200,000 GPUs in just 92 days. The company increased the previous 100,000 GPU maximum by adding 50,000 H100 and 50,000 H200 GPUs, bringing the total to 200,000 GPUs.

Grok 2 will soon be open source

As for Grok 2, the xAI team mentioned during their live Q&A session that it will be open-sourced in the next few months. On the topic of open source, many are pointing out that the popular AI model DeepSeek managed to outperform the competition without relying on a supercomputer with 200,000 Nvidia GPUs.

“Our general approach is that we will open-source the last version when the next version is fully out. When Grok 3 is mature and stable which is probably within a few months, then we will open-source Grok 2”

Elon Musk, Grok 3 launch Q&A

While DeepSeek might fall behind in the benchmarks shared by xAI, it's important to note that it was DeepSeek V3 in the comparison, not the more powerful R1 model. So, despite promoting Grok 3 as being capable of “solving physics’ greatest mysteries,” not everyone is convinced enough to try an AI model locked behind a paywall and one that doesn't work on a consumer GPU like DeepSeek does. On top of that, DeepSeek is already open source, and one expert has already claimed that it will help bring “the end of closed-source AI“.

About the Author

Hassam Nasir

Hassam boasts over seven years of professional experience as a dedicated PC hardware reviewer and writer.