From Kimi to DeepSeek and Alibaba... Chinese AI Models Launch an All-Out Offensive

Alibaba Officially Launches Qwen 3.8 Max
China Moves to Segment the AI Market
From Frontier Models to Ultra-Low-Cost Solutions

Chinese artificial intelligence (AI) companies have been ramping up their offensive by successively unveiling next-generation AI models over the past month. Starting with Moonshot AI’s Kimi K3, followed by DeepSeek’s V4-Flash and Alibaba’s Qwen 3.8 Max, each model touts unique strengths, reflecting the growing diversity within the Chinese AI ecosystem.

On August 4, Alibaba officially released its largest AI model to date, Qwen 3.8 Max. Instead of activating the entire model for every user request, Alibaba has reduced response latency by distributing tasks to specialized sub-modules. The model supports a multi-modal function that can process up to 1 million tokens of context as well as images and videos.

Although it is similar in scale to the Kimi K3, which was launched last month, Qwen 3.8 Max targets the market at a lower price point. The Qwen 3.8 Max model is based on a mixture-of-experts architecture with 2.4 trillion parameters, while Moonshot’s Kimi K3 has 2.8 trillion parameters. For reference, Qwen 3.8 Max is priced at $2 (about 3,000 won) for 1 million input tokens and $6 (about 8,600 won) for output tokens, making it less expensive than Kimi K3, which costs $3 (about 4,300 won) for input and $15 (about 21,000 won) for output tokens.

On August 1, DeepSeek unveiled V4-Flash. While its absolute performance is lower than top-tier models, operating costs have been significantly reduced. According to AI performance analytics firm Artificial Analysis, V4-Flash scored 50 points on the overall performance metric “Intelligence Index,” whereas Anthropic’s Claude Opus 5, Claude Fable 5, and OpenAI’s GPT-5.6 each scored 59 points. V4-Flash costs $0.14 (about 200 won) per million input tokens and $0.28 (about 400 won) per million output tokens, making its operating cost just one-105th that of Anthropic’s Claude Fable 5. Even compared to other Chinese models, V4-Flash is remarkably inexpensive.

Chinese AI models are carving out distinct segments of the market by leveraging varying strengths such as performance and price. By establishing differentiated positions—from frontier models targeting the highest performance, to general-purpose and ultra-low-cost models—they are broadening the AI ecosystem.

This division of roles is expected to further enhance the competitiveness of the Chinese AI ecosystem. By expanding the range of options to suit different purposes, companies can attract a wider spectrum of users.

Jeffrey Ding, a researcher at the Stanford Institute for Human-Centered Artificial Intelligence (HAI), commented, “China’s AI competitiveness no longer relies solely on DeepSeek. More than a dozen companies, including Moonshot AI and Alibaba, are developing powerful open-weight models, resulting in a highly diverse AI ecosystem.”

This content was produced with the assistance of AI translation services.