China has rapidly narrowed the gap with the US in producing cutting-edge artificial intelligence (AI) models, according to research published by Stanford University in California.
When measured using major AI evaluating benchmarks, the performance difference of Chinese models compared with their US counterparts shrank from double digits in 2023 to near parity last year, according to the latest Artificial Intelligence Index report from the Stanford Institute for Human-Centred AI.
For example, the performance gap between Chinese and US models in MMLU, or measuring massive multitask language understanding, narrowed from 17.5 points two years ago to 0.3 by the end of 2024, underscoring the relentless drive by Chinese tech firms to push the frontiers of the technology.
China’s Big Tech firms – including Alibaba Group Holding, ByteDance and Tencent Holdings, as well as AI start-ups such as DeepSeek and Zhipu AI – were ranked in the global top 15 in terms of notable AI models released last year. They included Alibaba’s Qwen series and DeepSeek’s V3 foundational model and its R1 reasoning model. Alibaba owns the South China Morning Post.
“The launch of DeepSeek’s V3 model in December 2024 garnered significant attention, particularly because it achieved exceptionally high performance while requiring far fewer computational resources than many leading [AI models],” the Stanford report said.
