Sebastian Raschka • 6/18/2026

GLM-5.2 and IndexShare for Long-Context Sparse Attention

This article reviews GLM-5.2, a new open-weight model from Z.ai, highlighting its architecture based on Multi-head Latent Attention and DeepSeek Sparse Attention (DSA). The key innovation is IndexShare, a cross-layer reuse trick that reduces computational cost for 1M-token inference by running the sparse-attention indexer only every four layers. The article compares GLM-5.2's coding benchmark performance favorably against Claude Opus 4.8, noting a 68.8 vs 56.7 score on the Artificial Analysis Coding Index. It is a technical analysis of a recent AI model release, relevant to IT/technology.

0 comments

#Sparse Attention #Glm 52 #Indexshare