Short note on GLM-5.2, an open-weight GLM update that keeps the GLM-5 sparse MoE backbone and adds IndexShare for cheaper 1M-token DSA inference.

Sebastian Raschka's Blog offers insights, tutorials, and research updates on machine learning, deep learning, and artificial intelligence. Covering topics such as neural networks, data science, and Python programming, Sebastian Raschka's Blog provides resources for students, researchers, and practitioners in the field of AI. Developers can learn about  algorithms, research methodologies, and practical applications of machine learning through Raschka's blog posts and publications.

Sebastian Raschka

GLM-5.2 is a new open-weight model from Z.ai that builds on the GLM-5/5.1 architecture, retaining Multi-head Latent Attention and DeepSeek Sparse Attention (DSA). The key innovation is IndexShare, a cross-layer reuse mechanism that runs the sparse-attention top-k indexer only once every four layers instead of every layer, significantly reducing compute cost for 1M-token inference while preserving adaptive attention patterns.

GLM-5.2 and IndexShare for Long-Context Sparse Attention