GLM 5.2

GLM 5.2 is Z.ai's flagship model built specifically for long-horizon work, described as a substantial leap over its predecessor GLM 5.1 in sustaining quality across extended agent trajectories. Its design centers on making long-context interactions genuinely usable for coding agents, rather than merely accepting more tokens. The model is released under an MIT open-source license with weights hosted publicly on Hugging Face, reinforcing its positioning as a transparent, open-weights foundation for research and product work that needs long-running reasoning chains.

A key architectural contribution in GLM 5.2 is IndexShare, which reuses the same indexer across every four sparse attention layers, cutting per-token compute at long context lengths while preserving quality. The MTP layer is also refined for speculative decoding, raising acceptance length and improving inference efficiency. Coding capability is strengthened through configurable thinking effort levels, letting developers trade latency against reasoning depth depending on task complexity. Together these traits make GLM 5.2 well suited to long-running coding agents and tool-driven workflows where sustained context quality, flexible compute budgets, and open deployment matter more than peak single-turn latency.