On February 17 at 8 PM Pacific Time, Elon Musk’s AI company, xAI, officially launched its latest large language model—Grok 3—in Los Angeles, California. During the launch event, Musk claimed that Grok 3's reasoning capabilities are an order of magnitude ahead of its predecessor, Grok 2, and called it “the smartest AI on Earth.” He also emphasized that Grok 3 outperforms other leading AI models, including ChatGPT and DeepSeek.
Core Innovations of Grok 3
During Grok 3's development, the xAI team implemented several cutting-edge technologies to enhance the model’s performance and adaptability. One of the most notable innovations is the introduction of the Chain of Thought (CoT) reasoning mechanism. This enables Grok 3 to process complex tasks step by step, providing more coherent and logically rigorous responses.
Additionally, Grok 3 was extensively trained on synthetic data, a technique that not only improves training efficiency but also allows the model to develop self-correction and reinforcement learning capabilities. By repeatedly reviewing data and ensuring logical consistency, Grok 3 can identify and correct its own mistakes, continuously optimizing its performance.
Grok 3 Surpasses DeepSeek in Reasoning Ability
Musk emphasized that Grok 3's reasoning capabilities surpass all currently available AI models.
During the launch, xAI showcased Grok 3's outstanding performance across multiple benchmarks in mathematics, science, and coding:
Mathematics (AIME'24 test): Grok 3 scored 52 points, significantly surpassing DeepSeek-V3’s 39 points.
Science Knowledge (GPQA evaluation): Grok 3 achieved 75 points, outperforming DeepSeek-V3’s 65 points.
Programming (LCB Oct-Feb coding test): Grok 3 scored 57 points, far exceeding DeepSeek-V3’s 36 points.
These results highlight Grok 3’s exceptional ability in mathematics, science, and programming.
Powerful Computational and Creative Abilities
During the live, Grok 3 demonstrated its real-time problem-solving skills by tackling complex physics problems, plotting interplanetary trajectories, and generating animated code that simulates spacecraft traveling between Earth and Mars. Additionally, Grok 3 developed a Tetris-like game on the spot, showcasing its creativity in game development.
Agents: Competing with OpenAI and Other Rivals
The release of Grok 3 poses a strong challenge to existing AI models, including OpenAI’s ChatGPT, DeepSeek’s V3, Anthropic’s Claude, and Google’s Gemini. In addition to the Grok 3 pre-trained model and two inference models, xAI also introduced a new AI Agent called DeepSearch. This feature is seen as xAI’s response to OpenAI and Google’s AI-powered research tools.
Colossus supercomputer
To support Grok 3’s training, xAI doubled the capacity of its Colossus supercomputer cluster within just three months, deploying 200,000 NVIDIA H100 GPUs. Furthermore, xAI is in negotiations for a $5 billion deal with Dell to purchase AI-optimized servers powered by NVIDIA GB200 chips, ensuring sufficient computing power for Grok 3 and future AI applications. These infrastructure investments demonstrate xAI’s technological strength, earning Musk’s signature phrase: “Big moves bring big miracles.”
User Experience and Future Plans
The beta version of Grok 3 is now available to Premium Plus users on X (formerly Twitter), with other users able to access it via a Super Grok subscription. Musk stated that the best version of Grok 3 will be available in about a week, featuring voice interaction capabilities for a more natural conversational experience.
When asked about potential open-sourcing, xAI stated that earlier versions of Grok would be open-sourced once Grok 3 reaches maturity, promoting knowledge sharing and AI community development.
Who Were the Two Prominent Chinese-face Scientists ?
At the center of the live were two notable Chinese AI scientists: Jimmy Ba (second from the left) and Yuhuai "Tony" Wu (third from the left), both founding members of xAI.
Jimmy Ba: From Academic Rising Star to xAI Core Researcher
Jimmy Ba completed his bachelor’s to PhD at the University of Toronto under Geoffrey Hinton, known as the “Godfather of AI.” He made significant breakthroughs in deep learning, most notably by developing the Adam optimization algorithm, now a standard method for training deep learning models. His contributions earned him the 2023 Sloan Research Fellowship, a prestigious award for outstanding early-career scientists. After his academic success, Jimmy Ba joined xAI as a core researcher.
Yuhuai "Tony" Wu: From Research to Industry
Tony Wu earned his PhD in Machine Learning from the University of Toronto in 2021 and interned at DeepMind and OpenAI. After completing his PhD, he pursued postdoctoral research at Stanford University and worked at Google’s N2Formal team, developing AI systems for automated mathematical reasoning. At Google, Wu collaborated closely with Christian Szegedy, advancing AI applications in mathematics.
How They Joined Forces with Musk
Interestingly, during his PhD, Tony Wu was a student of Jimmy Ba, laying the groundwork for their collaboration at xAI. Moreover, Christian Szegedy, Wu’s former supervisor at Google, also joined xAI, forming a strong connection that facilitated their partnership with Musk.
Musk has long been active in the AI space. In 2015, he co-founded OpenAI, but later parted ways due to differences in vision. In 2023, he launched xAI to explore the frontiers of AI development. Recognizing Jimmy Ba and Tony Wu’s expertise, Musk personally invited them to help shape xAI’s future.
Musk’s Grand Vision: Not Just Competing, But Changing the Game
Musk’s ambition in AI extends beyond competition—he aims to redefine the rules. He believes that the key to technological dominance lies in both the speed of innovation and the acceleration of that speed. This means not only achieving breakthroughs quickly but also building a self-sustaining system that continually accelerates innovation. The rapid evolution of the Grok series exemplifies this strategy.
For Musk, OpenAI was an experimental ideal, but xAI is the future he wants to control. He refuses to follow trends; instead, he seeks to lead them. His goal is not just to create the most advanced AI but to outpace competitors at an exponential rate. This is not merely a technological race—it’s a battle over who will truly lead humanity into the AI era. In this race, Musk’s objective is not to catch up, but to surpass—at an exponential scale.