DeepSeek upgrades AI model to challenge western tech giants

Chinese artificial intelligence startup DeepSeek has released a significant upgrade to its large language model, escalating competition in the global AI technology market.

The new model, DeepSeek-V3-0324, was launched on the Hugging Face AI development platform, demonstrating substantial improvements across multiple performance benchmarks. The company highlighted significant enhancements in several key areas, including reasoning capabilities, coding proficiency, and language-specific functionalities.

Specific improvements include advanced front-end web development capabilities, enhanced Chinese writing proficiency with features like “interactive rewriting,” and improved Chinese search capabilities such as “enhanced report analysis”. The model showcases notable upgrades in benchmark tests, positioning it as a more sophisticated AI solution.

Founded in 2023 by entrepreneur Liang Wenfeng, DeepSeek has rapidly emerged as a formidable challenger to established US technology companies. The startup previously claimed its V3 model, released in December 2024, was three times faster than its V2 predecessor and ranked among the top 10 in the Chatbot Arena performance platform hosted by the University of California, Berkeley.

The latest model represents a continuation of DeepSeek’s aggressive development strategy. The company previously launched its R1 advanced reasoning model in January, which achieved performance comparable to OpenAI’s offerings and surged to the top of iPhone download charts.

Notably, DeepSeek has distinguished itself through significantly lower development costs. The company stated that training its latest model cost approximately 5.6 million dollars, substantially less than the estimated 100 million to 1 billion dollars spent by other AI developers.

The model upgrade has not gone unnoticed by industry leaders. OpenAI’s chief executive officer Sam Altman previously praised DeepSeek’s products, describing the R1 model as “impressive … particularly around what [DeepSeek is] able to deliver for the price”.

However, the company faces potential regulatory challenges in the United States. A bipartisan bill was introduced in February to ban DeepSeek from federal devices, and some US agencies, including the Navy, Defence Department, and NASA, have already restricted access to the company’s products.

Despite these challenges, DeepSeek continues to make significant strides in AI development. The new V3-0324 model demonstrates the company’s commitment to improving AI capabilities while maintaining cost-effectiveness, positioning itself as a competitive alternative to established US tech giants.