精选理由
想了解阿里巴巴如何降低大模型训练成本的朋友，Qwen3.8-Flash值得一试，参数少，成本更低。
阿里巴巴Qwen团队发布Qwen3.8-Flash，Qwen4架构的多模态MoE预览，激活参数6B，训练成本仅为Qwen3.7-Plus的九分之一。
原文 · Pandaily
Alibaba Open-Sources Qwen3.8-Flash with 6B Active Parameters at One-Ninth Training Cost
Alibaba's Qwen team released Qwen3.8-Flash, a multimodal MoE preview of the Qwen4 architecture with only 6B activated parameters and a training cost one-ninth that of Qwen3.7-Plus.