DeepSeek’s AI Model R1-0528 Nears OpenAI’s Performance Tier

DeepSeek’s upgraded R1-0528 model now rivals leading AI systems from OpenAI and Google, marking a significant leap in China’s AI development. The model demonstrates enhanced reasoning, creative writing, and coding capabilities, with a 50% reduction in hallucinations—a critical improvement for reliability.

Post-training optimizations, fueled by additional computing power, sharpened the model’s efficiency and accuracy. On LiveCodeBench, R1-0528 trails only OpenAI’s o4-mini and o3, signaling competitive momentum. "DeepSeek isn’t just catching up; it’s contesting the frontier," noted Adina Yakefu of Hugging Face, highlighting strides in inference and hallucination mitigation.

The progress unfolds despite U.S. export restrictions on advanced chips to China, underscoring the resilience of Chinese tech firms.