Record Launch on Domestic Chips
China’s Zhipu AI officially launched its open-weight model GLM-5.3-Flash, previously code-named Ox Alpha. Specifically, the system ran entirely on a massive cluster of 100,000 domestically produced chips during a high-profile stealth trial. Consequently, Zhipu’s shares surged more than 12 percent in Hong Kong following the announcement. The model processed over 62 trillion tokens prior to release, setting a historic record on the OpenRouter marketplace. Furthermore, this deployment marks a crucial test of China’s ability to handle large-scale AI inference workloads entirely on home-grown hardware.
Architecture and Hardware Optimization
To overcome the lower memory bandwidth of domestic chips compared to Nvidia GPUs, Zhipu developed a specialized inference engine. This engine splits processing stages into independently managed computing pools. As a result, the firm tripled its end-to-end performance and brought hardware efficiency on par with mainstream Nvidia accelerators. While Zhipu did not name specific chip suppliers, domestic developers like Cambricon and Moore Threads announced day-zero compatibility. Additionally, the model features 320 billion total parameters while activating only 18 billion per request to reduce computing overhead.
Performance Benchmarks and Market Impact
Benchmarking firm Artificial Analysis ranked GLM-5.3-Flash tenth globally on its Intelligence Index and third among open-weight models. To attract international developers, Zhipu is offering aggressive pricing at a fraction of the cost of competing systems like Anthropic’s Opus 4.8. However, early developer feedback remains somewhat mixed, noting occasional hallucinations and slower output speeds alongside strong coding abilities. Meanwhile, the launch coincides with intensifying competition across China’s open-source ecosystem, as rival Alibaba recently introduced its own efficient Qwen3.8-Flash-Next model.
Reference
Chang, M., & Chang, M. (2026b, agosto 27). Zhipu AI shares jump as viral Ox Alpha model revealed as GLM-5.3-Flash on Chinese chips. South China Morning Post. https://www.scmp.com/tech/big-tech/article/3365433/zhipu-ai-shares-jump-viral-ox-alpha-model-revealed-glm-53-flash-chinese-chips?share=01Y8dt0MMQVrLhT3Ae%2FYT283mIfoKjl%2BXVLw11R5bPR%2BBBsh%2BrgU8iOS0eIr7HtZVvU72iQYaD1Tm6K80cvTuPsBhdHX6H3gn4ZDHHs36BY%3D&utm_campaign=social_share