GLM 5.2 is Z.ai's fifth-generation open-weight large language model, built for long-context reasoning, coding agents, and enterprise AI workflows. It pairs a Mixture-of-Experts architecture with sparse attention and multi-token prediction. The model uses a Mixture-of-Experts transformer design where a router selects the most relevant expert networks for each token, so only a fraction of the total parameters activate per token. GLM 5.2's native configuration is reported to support a usable context window of around 1 million tokens, with some hosted providers exposing smaller windows, such as 256K tokens. The key mechanism is dynamic sparse attention, where the model uses an indexer to select the most relevant parts of the context before applying full attention. Z.ai's technical material points to an optimization that shares a lightweight indexer across groups of sparse attention layers. Z.ai describes GLM 5.2 as built for long-horizon tasks, tuned for work that takes many steps, many files, and many tool calls, such as repository migration, policy analysis, bug triage, or agentic devops.