Chinese AI models have lagged the US frontier by 7 months on average since 2023 (epoch.ai)

Recent analysis reveals that since 2023, Chinese AI models have consistently lagged behind their American counterparts, with an average delay of seven months in capability, as measured by the Epoch Capabilities Index (ECI). This gap ranges from four to fourteen months and is primarily attributed to the fact that most leading Chinese models are open-weight, while cutting-edge US models remain proprietary. The analysis not only highlights the competitive landscape of AI development but also underscores the disparity in access to resources and model sophistication between the two countries. The implications of this finding are profound for the AI/ML community, as it points to a significant divide in innovation and advancement. The ECI data indicates that no Chinese model has matched or exceeded the performance of leading US models since January 2023, with projections suggesting this trend will continue. For example, the first Chinese model expected to surpass GPT-4 did not emerge until May 2024, reflecting a critical juncture in the global AI race. As countries vie for leadership in AI technologies, understanding these gaps could inform future strategies, funding, and research directions that either nation might pursue to close the performance divide.