Alibaba launches more efficient Qwen3-Next AI model

Jinse Finance reported that Alibaba's Tongyi Qianwen has released the next-generation foundational model architecture, Qwen3-Next, and open-sourced the Qwen3-Next-80B-A3B series models based on this architecture. Compared to the MoE model structure of Qwen3, this architecture features the following core improvements: hybrid attention mechanism, high-sparsity MoE structure, a series of training-stable optimizations, and a multi-token prediction mechanism that enhances inference efficiency. Based on the Qwen3-Next model structure, Alibaba has trained the Qwen3-Next-80B-A3B-Base model, which has 80 billion parameters but only activates 3 billion parameters. This Base model achieves performance comparable to or slightly better than the Qwen3-32B dense model, while its training cost (GPU hours) is less than one-tenth of Qwen3-32B, and its inference throughput for contexts above 32k is more than ten times that of Qwen3-32B, achieving exceptional cost-effectiveness in both training and inference.