Zhipu AI Releases GLM-5.3-Flash with Open Weights
Zhipu AI has released GLM-5.3-Flash, a Mixture-of-Experts (MoE) language model featuring 320 billion total parameters and 18 billion active parameters. The model's weights have been made publicly available on Hugging Face. The release of a high-performance open-weights model of this scale provides researchers and developers with a powerful tool for local deployment, potentially reducing reliance on proprietary APIs. It also highlights the growing competitiveness of Chinese AI labs in producing cost-effective, frontier-class models. Despite being labeled a "Flash" model, its 320B total parameters mean it requires substantial hardware, such as at least 192GB of RAM, to run even at Q4 quantization. Zhipu AI claims the model outperforms GLM-5.2 at one-tenth of the cost while approaching Claude Opus 4.8 on coding and agentic benchmarks.
## BACKGROUND
Zhipu AI is an artificial intelligence company spun off from Tsinghua University's NLP lab, known for developing the General Language Model (GLM) architecture. A Mixture-of-Experts (MoE) architecture optimizes computational efficiency by routing inputs to specific "expert" subnetworks, allowing a model to have a massive total parameter count while only activating a fraction of them during inference.