GLM-5.3

Zhipu AI · open weight · glm-5.3

GLM-5.3 is Z.AI's open-weight flagship for complex coding, long-horizon agentic work, and cybersecurity tasks. It uses the same base model as GLM-5.2, with its capability gains coming from post-training. The Hugging Face checkpoint is a text-only 753B-parameter MoE (about 40B active) with 78 layers, 256 routed experts plus one shared expert, eight routed experts selected per token, DeepSeek Sparse Attention with IndexShare, and a 1,048,576-token context window. It supports up to 128K output tokens, function calling, structured output, context caching, and always-on thinking with low, high, and max reasoning-effort levels. The downloadable FP8 and BF16 weights are available under the custom GLM-5.3 License, which permits use, modification, redistribution, sale, deployment, and fine-tuning but requires Model-as-a-Service businesses above US$10B in annual revenue to pass a Z.AI security review before commercial use.