阿里正式发布新一代大模型Qwen3.5

Core Viewpoint - Alibaba's Qwen officially released Qwen3.5, introducing the first model in the Qwen3.5 series, Qwen3.5-397B-A17B, with open weight version [1] Group 1: Model Features - The model utilizes an innovative hybrid architecture combining Gated Delta Networks (linear attention) and Mixture of Experts (MoE) [1] - It achieves excellent inference efficiency with a total parameter count of 397 billion, activating only 17 billion parameters during each forward pass [1] - The design optimizes speed and cost while maintaining performance capabilities [1]