ISTANBUL

Chinese-based AI startup DeepSeek has released an updated model with improved programming capabilities to compete with leading US models like OpenAI.

The update, posted on the open-source AI community platform Hugging Face, tackles real-world challenges while setting new standards for accuracy and efficiency, the company said.

In January, DeepSeek's R1 model, which matched OpenAI’s top offerings, helped it become the most popular free app on Apple’s US App Store, triggering industry shock and a selloff on US markets.

The company spent just $6 million to train its V3 model, much less than OpenAI's $100 million for GPT-4, and used 10% of the computing power Meta used for Llama 3.1, making DeepSeek a major, cost-effective AI competitor.

DeepSeek’s models feature “open weights,” allowing use with some restrictions on modifications, unlike fully open-source AI.

The company recruits top AI experts from leading Chinese universities and specialists from various industries to enhance the intelligence and diversity of its models.

Chinese President Xi Jinping held a meeting with tech leaders and entrepreneurs, during which he encouraged business leaders to maximize their potential.

“It’s the right time for private enterprises and entrepreneurs to give full play to their potential,” he said.