Alibaba Unveils New AI Model, Claims It Outperforms Moonshot's Kimi K3

Alibaba has introduced its latest artificial intelligence model, Qwen3.8-Max, claiming it outperforms Moonshot AI's Kimi K3 on several key benchmarks, Andrew Nusca reported for Fortune Tech.

Alibaba has launched Qwen3.8-Max as it intensifies competition in China's fast-growing AI industry. [Photo: Qwen Cloud Facebook]

The launch marks another step in the intensifying competition among Chinese AI developers, where companies regularly release new models claiming improved performance.

Kimi K3, introduced on July 16, attracted global attention following the success of DeepSeek's R1 and quickly became one of China's leading large language models.

Alibaba said Qwen3.8-Max is optimized for autonomous coding, advanced research, and other long-duration reasoning tasks.

Like Kimi K3, the new model uses a mixture-of-experts (MoE) architecture capable of processing text, images, and video. According to Alibaba, the model can improve its performance through iterative feedback during complex tasks.

Qwen3.8-Max contains 2.4 trillion parameters, compared with 2.8 trillion for Kimi K3.

However, Alibaba said its model activates only 95 billion parameters during inference, compared with 104 billion for Kimi K3, helping reduce computing costs and response times.

Alibaba also highlighted its rapid development pace, noting that Qwen3.8-Max was released only two months after Qwen 3.7, while Moonshot took about a year to develop Kimi K3 following Kimi K2.