Z.ai Releases GLM 5.2 Open Weights Model With Post-Training Code

Chinese AI firm Z.ai released the GLM 5.2 model with open weights, making the code and training pipeline available to developers. The post-training release includes code for training using Megatron-LM, a widely adopted framework for large-scale distributed training, and SGLang, a high-throughput inference engine used for generating model rollouts.

The open-source availability of the post-training code allows researchers and builders to inspect, modify, and extend the alignment and reinforcement learning steps typically used to fine-tune models. The code is shared under the MIT license, granting commercial and academic users broad rights to adapt the model's training and rollout architecture for their own applications.