Japanese Media: China's High-Value AI Models Will Attract Japanese Enterprises and Developers
According to the Japanese media outlet ASCII, on August 3: On the 3rd, Alibaba unveiled its latest AI model, "Qwen 3.8-Max," with a total of 2.4 trillion parameters. It also plans to begin offering open-weight training in the future. As a result, the growth of AI models and cost reduction are accelerating in tandem.
Qwen 3.8-Max is Alibaba’s highest-performance AI model ever developed. Users can input images and videos, and process up to one million tokens simultaneously. Its goal is to autonomously execute large-scale data analysis and development, as well as multi-task AI agents. With 2.4 trillion parameters, it does not activate all parameters every time it answers a question. Instead, it employs a method known as "MoE" (Mixture of Experts), using only the necessary parts based on the content being processed—effectively utilizing around 95 billion parameters in practice.
Following the announcement of Alibaba’s new model, competition among AI companies has intensified further.
On July 17, Moonshot AI released "Kimi K3," featuring 2.8 trillion parameters. The model’s weights were made available on July 27.
Meanwhile, DeepSeek is strengthening its low-cost strategy. On July 31, DeepSeek officially launched "DeepSeek-V4-Flash" via API, featuring 284 billion parameters and 130 billion runtime capacity, released in public testing form. The API pricing is set extremely low: input costs $0.14 per million tokens (approximately 20–21 JPY), and output costs $0.28 per million tokens (approximately 39–42 JPY).
The real game-changer is the exceptional cost-performance ratio offered by Chinese AI models.
Among American developers, AI models produced in China are increasingly recognized as an affordable and practical option. As AI performance improves and AI agents enable parallel execution and scalability across multiple models, cost efficiency during usage will become even more critical. The rise of Chinese AI models—with their disruptive pricing strategies—could significantly influence model selection decisions by Japanese enterprises and developers.
Original article: toutiao.com/article/1872480768452608/
Disclaimer: The views expressed in this article are those of the author(s) alone.