Elon Musk’s xAI has introduced the new Grok 3 series of models, which includes a pre-trained model similar to GPT-4o, two reasoning models, and an AI assistant called Deep Search. The official launch of Grok 3 took place yesterday, on February 17, and it will gradually become available across all regions.

Grok 3 is powered by xAI’s Colossus supercomputer, which contains over 100,000 Nvidia GPU hours. This powerful system was built in just eight months.

The new, enhanced model has been trained on synthetic datasets using machine learning techniques, including reinforcement learning.

Reinforcement learning is considered an effective method for reducing incorrect or fabricated responses from AI models, also known as “hallucinations.” This approach also significantly shortens the training time of Large Language Models (LLMs).

Musk praised his new AI chatbot, calling it "terribly smart" and stated that it surpasses all previously released models.

“This might be the last time an AI is better than Grok,” Musk said at the World Government Summit in Dubai, emphasizing the advanced capabilities of the new model. Notably, reports suggest that Grok 3 will also feature text-to-video generation capabilities.

"We need to spend less money and get better results," Musk stated, seemingly making a direct reference to the global-scale corporate transformation in the AI sector.

With this latest release, competition in the already intense AI industry is expected to heat up even further.