Ox Alpha: "Pure Chinese Chips," 100x Cheaper AI Model

Ox Alpha, a new AI model, is powered entirely by Chinese chips. Z.ai founder Jie Tang announced that their GLM-5.3-Flash model, known as Ox Alpha, achieved a score of 57 on the Artificial Analysis Intelligence Index. This places it in the same range as advanced Western models, but at a price that is 100 times lower. Ox Alpha was a stealth release on OpenRouter, offering a 1-million-token context window and supporting text, image, and video input. It was available for free for about a week, processing 100 trillion tokens daily. Z.ai confirmed that Ox Alpha is GLM-5.3-Flash, a 320-billion-parameter hybrid model. It uses sparse activation, only activating 18 billion parameters per token, which allows for its aggressive pricing. This model reportedly outperforms its predecessor, GLM-5.2, and is significantly cheaper. The company states GLM-5.3-Flash ran on a cluster of domestically produced Chinese AI chips during its free week. This demonstrates a significant advancement in domestic silicon infrastructure. This development could reshape the global AI landscape and its accessibility.