1 Matching Annotations
  1. Last 7 days
  2. siliconangle.com siliconangle.com
    https://siliconangle.com/2026/08/26/z-ai-open-sources-ox-alpha-model-as-glm-5-3-flash/
    1
    1. fxp007 28 Aug 2026
      GLM-5.3-Flash features a mixture of experts architecture with 320 billion parameters. It activates 18 billion parameters to answer prompts.

      320B 总参只激活 18B，激活率不到 6%，比上一代激进得多。这个配比说明中国厂商已把稀疏度当成主要降本旋钮，而不是缩小模型。后果是部署门槛卡在显存容量而非算力，反而更依赖大内存硬件。

      moe sparsity glm
    Visit annotations in context

    Tags

    Annotators

    URL

    siliconangle.com/2026/08/26/z-ai-open-sources-ox-alpha-model-as-glm-5-3-flash/