Zai: GLM-5.3-Flash Confirmed as Ox Alpha Model

Zai confirmed its GLM-5.3-Flash as the viral Ox Alpha model, packing 320B parameters with 18B active per token through hybrid linear and sparse attention. The model leads GDPval AA v2 benchmarks at just $0.09 per task while running the entire free preview on China-made hardware via aggressive memory optimization. DeepLearning.AI highlighted the LLM architecture details in its latest Batch report, underscoring gains in AI model performance and real-world deployment efficiency.