On August 25, 2026, Z.ai (the GLM team) released two model repositories on Hugging Face: "zai-org/GLM-5.3-Flash" and "zai-org/GLM-5.3-Flash-BF16".

Both models utilize an image-text-to-text format, accepting images and text as input to produce text output, and support English and Chinese. Both share the tags transformers, safetensors, glm5_next, conversational, and endpoints_compatible, and reference a related arXiv paper (2602.15763). The license is MIT, and the region is listed as US.

While "GLM-5.3-Flash" is tagged with fp8 quantization and eval-results, these tags are not present for "GLM-5.3-Flash-BF16".


Sources: