AI News (2026/2/17): Qwen3.5 Team Officially Releases the Next-Generation Model
Executive Summary:
The Qwen3.5 team has officially released the next-generation Qwen3.5 series, with the flagship model Qwen3.5-397B-A17B featuring an innovative hybrid attention and sparse MoE architecture, significantly improving inference efficiency. This model competes with top-tier models like GPT-5.2, Claude 4.5, and Gemini 3 Pro in various cutting-edge benchmark tests, demonstrating comprehensive and leading performance.
通义千问团队正式发布新一代模型Qwen3.5
News Details
The Qwen team officially released the next-generation Qwen3.5 series on Monday, February 16. The flagship model, Qwen3.5-397B-A17B, is a native multimodal model that incorporates an innovative hybrid attention mechanism and a sparse MoE (Mixture of Experts) architecture. These technological advancements have significantly improved the model's inference efficiency, while also delivering excellent performance in various cutting-edge benchmark tests.
Key Points
Hybrid Attention Mechanism: Qwen3.5-397B-A17B introduces a hybrid attention mechanism that combines self-attention and cross-modal attention, enabling more efficient processing of multimodal data. This mechanism not only enhances the model's inference speed but also improves its ability to understand complex tasks.
Sparse MoE Architecture: The model adopts a sparse MoE architecture, which dynamically selects expert sub-networks to process input data, significantly reducing the consumption of computational resources. This architecture allows Qwen3.5-397B-A17B to maintain high performance while being more adaptable to different application scenarios.
Benchmark Test Performance: Qwen3.5-397B-A17B was tested against top-tier models like GPT-5.2, Claude 4.5, and Gemini 3 Pro in various cutting-edge benchmark tests, including instruction following, general intelligence agents, visual language, spatial intelligence, and video understanding. The results show that Qwen3.5-397B-A17B excels in multiple metrics, demonstrating its comprehensive and leading performance.
AI-ALL In-Depth Commentary
The release of the Qwen3.5 series models by the Qwen team, particularly the flagship Qwen3.5-397B-A17B, marks another significant advancement in multimodal AI technology. The combination of the hybrid attention mechanism and the sparse MoE architecture not only enhances the model's inference efficiency but also boosts its performance in complex tasks. This technological breakthrough has far-reaching implications for the AI ecosystem, especially in applications that require efficient processing of multimodal data, such as intelligent customer service, content generation, and video analysis. Additionally, the outstanding performance of Qwen3.5-397B-A17B in multiple benchmark tests further solidifies its competitive position in the AI field, providing developers with more choices and possibilities.
