zai-org/GLM-5.3-Flash · Hugging Face

Read https://huggingface.co/zai-org/GLM-5.3-Flash for the benchmark figures on models/glm-5-3-flash. As recorded by the reader: Only one allowed-list benchmark is published for GLM-5.3-Flash: Terminal-Bench 2.1 = 84.3 (Claude Code 2.1.207 harness, temp=1.0, top_p=1). It appears in the official Z.ai GLM-5.3-Flash release blog table and the same 84.3 is printed in the benchmark chart embedded in the official HF model card (read from the card image). All other published numbers are non-allowed benchmarks: DeepSWE v1.1 63.4, AutomationBench v1.0.6 48.8, Agents' Last Exam 26.3, HLE w/ tools 55.3, GDPval-AA v2 1773, Toolathlon Verified 78.4, NL2Repo 56.3, plus vision benchmarks. The blog's base-model table reports MMLU 88.1 for GLM-5.3-Flash-Base (pre-trained checkpoint, explicitly not the released instruct model), so it is not reported as this model's MMLU.