Musk Officially Unveils Grok-3

Elon Musk has introduced Grok-3, the latest version of xAI’s chatbot, which, according to developers, is ten times smarter than Grok-2.

In benchmark tests, the model outperformed GPT-4o, Claude 3.5 Sonnet, Gemini-2 Pro, and DeepSeek-V3 in mathematics, programming, and scientific computing.

According to Musk, Grok-3 has advanced reasoning capabilities and will continue to improve. During the presentation, the chatbot calculated an optimal launch window for a Mars mission and return trip to Earth, factoring in all necessary technical parameters.

Grok-3 was trained on the Colossus giga-computer, which utilizes 200,000 Nvidia H100 chips.

One of the key innovations is DeepSearch—a mode where the AI scans the entire accessible internet to find accurate and reliable answers. If information is insufficient, the model generates reasoned predictions, such as a year-ahead weather forecast or the optimal strategy for a new video game.

Additionally, Grok-3 can generate games in real time. During the demonstration, it created a hybrid of Tetris and match-3 puzzles, which Musk described as a near-autonomous AI game studio.

xAI clarified that Grok-3’s reasoning model is still in beta testing and continues training. A lighter version of the model is also available for less resource-intensive tasks.

The planned voice mode was not included in the current version, as it is still unstable. Developers expect it to launch within a week.