Glm 5.2

GLM 5.2 was introduced as Z.ai's flagship model built for long-horizon tasks, marking a substantial leap over its predecessor GLM 5.1. Its defining trait is a solid 1M-token context window engineered to sustain quality across extended, messy coding-agent trajectories rather than simply accepting more tokens. The launch post frames long-context engineering as a practical requirement for agentic workflows, positioning GLM 5.2 for sustained software engineering and multi-step reasoning sessions where shorter-context models tend to drift.

For coding workloads, GLM 5.2 introduces a flexible effort mode that exposes multiple thinking effort levels, letting callers trade latency against reasoning depth, alongside an architectural refinement the team calls IndexShare that reuses a shared indexer across four sparse attention layers to cut per-token compute at long contexts. The model is released as open weights under an MIT license with no regional limits, and its successor GLM 5.3 reuses the same base, meaning GLM 5.2 is the foundation model that later post-training iterations build on. This combination of long context, adjustable reasoning effort, and permissive open weights makes GLM 5.2 well suited to teams running local or hosted coding agents that need sustained attention over large codebases.