Chinese AI lab Z.ai confirmed Wednesday that Ox Alpha — the anonymous model that took over usage charts last week — is its new GLM-5.3-Flash. The model ran for a week on 100,000 Chinese-made chips, serving 62 trillion tokens before the official launch. No Nvidia hardware.
Ox Alpha hit number one on OpenRouter by usage, doubling the second-place model. It processed 11 trillion tokens in its first three days, the platform's biggest launch ever. Z.ai says the entire preview ran on domestically produced Chinese AI chips.
The model itself
GLM-5.3-Flash is a 320 billion parameter model with 18 billion active. It's open-weight, MIT-licensed, multimodal with a 1 million token context. API pricing is $0.15 input and $0.50 output per million tokens — about one-tenth the cost of comparable models. Artificial Analysis ranks it 10th globally, ahead of DeepSeek V4 Pro Max.
Why it matters
China has been cut off from advanced Nvidia chips since 2022. The assumption was that Chinese labs couldn't compete at scale without them. This week changed that.
Z.ai claims its Chinese chip setup now serves tokens about as cheaply as Nvidia hardware. If true, it removes one of China's key AI bottlenecks. The company's Hong Kong-listed shares jumped 12% Thursday. The question isn't just whether the model is good. It's whether China just solved its hardware problem while the West was busy assuming it couldn't.