阿里公布全模态模型新进展，Qwen4和下代视频模型均在训练中

AI 导读

阿里公布大模型最新进展，Qwen4及下一代视频模型正在训练，多模态模型性能提升，RSI技术实现模型自我进化，开源模型全球影响力显著。

Alibaba unveils advances in large models, including Qwen4 and next-gen video models in training, enhanced multimodal capabilities, and RSI-driven self-evolution, with open-source models gaining global traction.

重点速览

• Qwen4及下一代视频模型参数量将扩展至5-10万亿，生成力与控制力显著提升 Qwen4 and next-gen video models will expand parameters to 5-10T, enhancing generation and control capabilities
• Qwen3.8-Max通过RSI技术实现零人工参与的自主迭代，性能提升12.5% Qwen3.8-Max achieves 12.5% performance boost via RSI-driven zero-human iteration
• 多模态模型覆盖图像/音频/视频领域，Wan3.0文生视频双榜全球第一 Multimodal models cover image/audio/video domains, with Wan3.0 leading global video generation
• 开源模型Qwen3.8下载量超5600万，衍生模型超1900个，成为Hugging Face最受欢迎模型 Qwen3.8 open-source models exceed 56M downloads, spawning over 1900 derivatives

一句话：阿里通过技术创新与开源策略，巩固其在大模型领域的全球领先地位。 / Alibaba strengthens its global leadership in large models through technological innovation and open-source strategies.

推荐理由 展示阿里在AI领域的持续创新与实际应用案例，对行业技术演进有重要启发。