xAI, the startup led by Elon Musk that raised $6 billion in December, has a new AI model that it claims is better than AI created by DeepSeek and ChatGPT-maker OpenAI.
In a live-streamed event on X on Monday that has been viewed over six million times at the time of writing, Musk and three xAI engineers revealed Grok 3, the startup’s latest AI model. They claimed Grok 3 had higher scores on math, science, and coding benchmark tests than OpenAI’s GPT-4o, DeepSeek’s V3, and Google’s Gemini AI.
Related: Elon Musk’s xAI Is Reportedly Set to Hire Thousands of ‘AI Tutors’ With Pay Up to $65 an Hour
They also said Grok 3 was a step up in sheer power from xAI’s previous model Grok 2, released in August. The latest version has more than 10 times the computational power of Grok 2, greater accuracy, and a bigger capacity for large datasets.
“The word Grok [means] to fully and profoundly understand something,” Musk said on the livestream, noting that the word came from the 1961 novel “Stranger in a Strange Land” by American author Robert Heinlein. He added later in the livestream that “if you’re using Grok 3, you may notice improvements almost every day because we’re continuously improving the model.”
Animated 3D plot of a spacecraft launch from Earth to Mars and back. Credit: xAI
xAI engineers demonstrated how Grok 3 could be used to create code for an animated 3D plot of a spacecraft launch that started on Earth, landed on Mars, and came back to Earth.
The engineers also asked Grok to combine two games, Tetris and Bejeweled, into one game. The result, which the engineers played on the livestream, was similar to Tetris with shapes inching down the screen but had the rules of Bejeweled with multicolored blocks that disappeared if there were three in a row.
Related: Google’s CEO Praised AI Rival DeepSeek This Week for Its ‘Very Good Work.’ Here’s Why.
Musk said that any AI could find examples of Tetris or Bejeweled online and duplicate them, but Grok 3 took it one step further.
“What’s interesting here is it [Grok 3] achieved a creative solution combining two games that actually works and is a good game,” Musk noted. “We’re seeing the beginnings of creativity.”
Tetris-Bejeweled mashup game in the background. Credit: xAI
The researchers said they only trained Grok 3’s reasoning abilities on math problems and competitive coding problems, but they observed that Grok 3 could apply what it learned to a variety of use cases, including reasoning through making games.
xAI isn’t the only major AI startup to release advanced AI this year. Last month, OpenAI released the o3-mini, its most cost-effective yet powerful model yet, while DeepSeek came out with R1, a disruptive AI model with cutting-edge performance on a less than $6 million budget.
Grok 3 is currently available for Premium+ X subscribers paying $22 a month.
Watch the event, here:
https://t.co/hEfQ31gANQ
— xAI (@xai) February 18, 2025