2025 was marked by rapid developments in artificial intelligence (AI), including the release of new models, the advancement of agentic AI systems, and the expansion of multimodal capabilities. From the beginning of the year to the end, competition intensified between leading companies in the United States and China, setting new benchmarks.
The progress of the Chinese company DeepSeek marked the beginning of the year. Released in January, the open-source DeepSeek R1 model, which cost only $ 6 million to train, outperformed the leading models from OpenAI and Google in mathematics and programming tests. In December, the company introduced DeepSeek-V3.2 and V3.2-Special versions, which are competitive with Gemini 3.0-Pro and are distinguished by high efficiency in performing long-term logic and agent tasks.
The Grok chatbot from xAI also made significant progress. The Grok 3 model, released on February 17, was trained on the Colossus supercomputer using nearly 200,000 GPUs, providing 10 times the computing power of its predecessor. It led the way in math, science, and programming benchmarks and was temporarily made available for free to Premium+ subscribers and regular users. In July, Grok 4 was introduced, including the Heavy version, which is considered one of the most intelligent models in the world. On November 17, Grok 4.1 was introduced, with improved creative, emotional, and collaborative capabilities.
