On June 13, 2026, Beijing-based Z.ai, formerly known as Zhipu AI, released GLM-5.2, the newest flagship in its GLM-5 family, aimed at agentic coding and long-horizon tasks. Per the Hugging Face model card, the model is a Mixture-of-Experts architecture with 753 billion total parameters and a usable 1-million-token context window that, in the card’s words, “stably sustains long-horizon work,” along with selectable thinking-effort levels to balance performance against latency.
The model card publishes GLM-5.2 under the permissive MIT license, with no regional restrictions, and provides FP8 and other quantized variants. That is an unusually open stance for a frontier-class model: MIT licensing allows commercial use, modification, and self-hosting with minimal obligations.
For a business reader, the significance is the combination of capability, openness, and cost. A frontier-class coding model that can be self-hosted under MIT terms, with a million-token context, sharpens the open-versus-closed decision for teams weighing whether to depend on a metered API from a Western lab or run a comparable model on their own infrastructure.