GLM-5.2 is Z.ai's flagship model purpose-built for long-horizon tasks, representing a substantial leap over its predecessor GLM-5.1. It ships as a fully open-weights release under an MIT license with no regional access limits, initially appearing to coding-plan subscribers before the full weights became publicly downloadable. The model is structured as a 753-billion-parameter Mixture-of-Experts architecture with around 40 billion active parameters, and it sustains a solid 1-million-token context window designed for stable operation across extended agent trajectories rather than merely accepting very long inputs.
Z.ai highlights several technical improvements in GLM-5.2, including an architectural feature called IndexShare that reuses the same indexer across every four sparse attention layers to cut per-token compute at long context lengths, alongside refinements to the model's multi-token prediction layer that boost speculative decoding acceptance. The release emphasizes advanced coding capabilities with configurable thinking effort levels so developers can trade latency for performance, and independent coverage notes the model ranks competitively on agentic front-end coding leaderboards. GLM-5.2 is particularly well suited to project-level software engineering, long-running coding agents that must retain engineering context through multi-step workflows, and complex automation pipelines that demand consistent tool use over extended sessions.
