GLM 5.2 becomes the open-weight model agentic coding teams actually trust
Zhipu's 753B MoE model, MIT-licensed with roughly 40B active parameters per token, is emerging as the default self-hosted choice for coding agents.
Open-weight models have had a strong 2026, but GLM 5.2 from Zhipu is the one developer communities keep converging on for a specific, unglamorous reason: reliability under long agentic tool-use loops, not raw benchmark position.
The model is a 753-billion-parameter mixture-of-experts design with about 40 billion parameters active per token, released under an MIT license — meaning no restrictions on commercial fine-tuning or redistribution. That combination of scale, efficiency and permissive licensing has made it a popular base for teams building self-hosted coding agents who don't want to route sensitive codebases through a closed API.
It's a useful data point on where open-weight competition is actually landing in 2026: not in chasing frontier benchmark scores, but in being 'boring' enough — consistent, inspectable, self-hostable — to trust with production agent workloads.