阿里 Qwen4 和下代视频模型训练中，Qwen4.5 后模型将扩展至 5-10T 参数

AI Digest

阿里公布Qwen4及下一代视频模型进展，参数扩展至5-10T，RSI技术提升模型能力，视频模型将11月发布，具备更长、更智能特性。

Alibaba reveals Qwen4 and next-gen video model progress, expanding parameters to 5-10T, RSI technology enhances capabilities, with a new video model set for November release, offering longer, more controllable, and intelligent features.

Key points

• Qwen4.5及后续版本参数规模将达5-10T，显著提升模型能力 Qwen4.5 and later versions will expand parameters to 5-10T, significantly enhancing model capabilities
• RSI技术实现模型自我训练与推理优化，突破传统训练模式 RSI technology enables self-training and inference optimization, breaking traditional training models
• 下一代视频模型11月发布，支持长时一致性与导演级创作 Next-gen video model to launch in November, supporting long-term consistency and director-level creation
• Qwen3.8-Max在零人工参与下完成33轮迭代，性能达国际一流 Qwen3.8-Max completes 33 iterations without human intervention, achieving top-tier performance
• AI手机全栈方案发布，推动大模型终端化应用 AI phone full-stack solution released, advancing large model terminalization

Takeaway: 阿里大模型技术持续突破，参数规模与智能能力显著提升。 / Alibaba's large model technology continues to break through, with significant improvements in parameter scale and intelligence.

Why it matters 展示阿里大模型技术前沿进展，对AI行业有重要参考价值。