Despite calls from some AI companies, including Anthropic, to slow down development due to safety concerns, the competition between China and the United States in AI development shows no signs of cooling.
According to the Nihon Keizai Shimbun on October 7, ten major Chinese AI companies, including DeepMind, Alibaba, Tencent, and Xiaomi, launched 16 AI models in September alone. This count includes only newly developed models for image and audio generation, excluding modified versions of existing systems.
Statistics from the AI benchmarking site BenchLM confirm the increasing trend of Chinese AI models. Eight major Chinese AI developers tracked by BenchLM released 20 large language models (LLMs) from July to September, a significant increase from just two in the first quarter and 17 in the second quarter.
The pace of AI model releases is also accelerating in the United States. From January to September, U.S. developers introduced a total of 59 models, approximately 50% more than their Chinese counterparts. The breakdown includes 10 models in the first quarter, 15 in the second, and 34 in the third quarter.
Notably, Chinese AI models are leading in actual usage, emphasizing low cost and flexibility. According to the AI model brokerage platform OpenRouter, Chinese AI models processed 62.22 trillion tokens during the week of September 21-27, compared to 14.2 trillion tokens for U.S. models. Token processing volume serves as an indicator of how extensively AI models are utilized.
Hu Yanping, a professor at Shanghai University of Finance and Economics, analyzed in an interview with China Daily on October 7 that the increase in usage of Chinese AI models is due to the variety of available options, competitive pricing, the abundance of open-source models, and rapid development speed.
As AI agents become more widespread, these characteristics are becoming increasingly important. AI agents perform multiple tasks, such as writing code, searching databases, and calling external programs, which leads to significantly higher token usage compared to standard chatbots. For businesses, the incentive to choose lower-cost models increases if performance is comparable.
Consequently, companies are prioritizing cost-performance ratios and the potential for utilizing AI agents over simple performance metrics when selecting AI models.
Chinese companies are also reducing the time it takes to move models from testing to official release. When new models are introduced, existing users are quickly transitioned to the new services, and the models are distributed across multiple platforms to ensure that technological improvements translate into increased usage.
To close the technological gap with the United States, Chinese companies are accelerating the development of large-scale models. Alibaba plans to increase the parameters of its AI model Qwen, currently at 240 billion, to between 500 billion and 1 trillion in the future.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.