27 August 2026updated 30 August
Z.ai releases open-source GLM-5.3 coding model and cheaper variant
First reported
Deep Learning Weekly ran this on , a day before the other 3 sources picked it up.
- Z.ai open-sourced GLM-5.3, a model trained to find bugs in code. It identified 2,436 bugs across 269 real open-source projects.
- Z.ai also released GLM-5.3-Flash, a smaller version of the same model designed to cost roughly one-tenth as much to run.
- GLM-5.3-Flash can process 1 million tokens, a context window [how much text it can hold at once], and handle images alongside text.
Where they differ
The Neuronemphasized the bug-finding capability and results.
Deep Learning Weeklyfocused on the cheaper Flash variant's technical architecture and cost advantage. Both releases happened together but each newsletter led with different aspects of the announcement.
What each one reported
Z.ai open-sourced GLM-5.3 after post-training improved coding and cybersecurity performance, with the model finding 2,436 bugs across 269 open-source projects.
Z.ai released GLM-5.3-Flash, a 320B/18B MoE model with hybrid sparse-plus-linear attention and 1M-token multimodal context, positioning it as a cost-effective alternative to the full GLM-5.3 model.
Tencent released Hy4 Preview, a new open weight text LLM with 770B total parameters, 49B active parameters, and a 1M token context window. The model is available on Hugging Face.
Z.ai launched GLM-5.3-Flash, a 320B-parameter model with only 18B active parameters, achieving frontier-level coding performance at lower inference costs. It scored 63.4 on DeepSWE and reached near-parity with Claude Opus on several coding tests while supporting up to 1M-token contexts.