Chinese AI Closing the Gap with US, What Competitive Advantage Remains for the West?

Chinese AI models are closing the gap with top-tier US models in terms of performance. According to the monthly technical analysis report "Frontier Radar" Issue 4, China's large language models "Kimi K3" and "GLM-5.3" have reached performance levels comparable to the most advanced US models at the present time.

Until now, the forefront of AI development has been led by US research institutions and companies such as OpenAI and Anthropic. Western nations have maintained their competitive advantage in the AI race by keeping a lead in model performance. However, this analysis suggests that this underlying assumption may be faltering.

Some Western research institutions cite "distillation" as one factor behind China's rapid model performance improvements. Distillation is a technique where the output of an already-developed high-performance model is used as training data to efficiently develop smaller models. Since performance can be enhanced by "transferring" knowledge from existing high-performance models, it is believed that models can approach comparable capabilities with fewer resources than developing from scratch. According to the report, concrete evidence exists that Chinese models have leveraged this technique.

However, the report raises a more fundamental question. Regardless of whether distillation was used, the conclusion is that "performance differences in models can no longer serve as a competitive defense line." In other words, we are entering an era where possessing a high-performance model alone cannot maintain competitive advantage. The report then places at the center of its analysis the question of what could become the source of future competitive advantage.

This question presents a structural challenge to the entire AI industry. If models themselves move away from being the axis of differentiation, competitive focus is likely to shift to peripheral areas such as data, infrastructure, implementation capability in application domains, or regulatory environments and partnerships with corporations and governments. This can be seen as a questioning of the very premise under which performance benchmarking has functioned as an industry-standard evaluation metric.

International competition in AI models has largely been framed around the question "who can build the smartest model." However, going forward, the question "where do we compete in an era where intelligence itself cannot be differentiated" is likely to become more critical to the strategies of individual nations and enterprises. This shift in perspective can be expected to influence research and development investment direction and business strategy, and deeper discussion on this issue is anticipated.