The Beijing company Z.ai Ox Alpha GLM-5.3-Flash has been confirmed as one and the same model: the firm revealed this week that its anonymously released Ox Alpha was in fact GLM-5.3-Flash, its latest open-weight large language model, which it had been quietly testing on OpenRouter and OpenCode under a pseudonym to gather real-world user feedback before a formal release.
The disclosure has thrust Z.ai, also known in China as Zhipu AI, into the spotlight at a moment when competition among Chinese AI developers is intensifying. The company says it was founded in 2019, drawing on technology developed at Tsinghua University in Beijing, and has been building out its General Language Model (GLM) family of large language models ever since. Its chatbot, ChatGLM, launched in 2023.
How Z.ai Ox Alpha GLM-5.3-Flash Fared Against the Competition
The model’s anonymous run on public leaderboards produced results that drew attention. On the Artificial Analysis Intelligence Index, GLM-5.3-Flash places 10th overall, putting it ahead of DeepSeek V4 Pro Max, according to Quartz. That ranking, achieved while running on Chinese AI chips rather than the advanced semiconductors Washington has restricted China from accessing, added a layer of strategic interest to what began as a stealth product test.
Z.ai said the Chinese chips it used had limitations in memory capacity and bandwidth. To compensate, its engineers built a dedicated inference engine on top of the SGLang framework, achieving a 3x improvement in serving performance over its initial baseline on the same hardware, Quartz reported. The optimisation effort, combining software tuning with careful management of how workloads were distributed across the chips, allowed the company to serve the Ox Alpha traffic at competitive speed despite the hardware constraints.
GLM-5.3-Flash is a multimodal model, capable of processing text, images and video. Its weights are publicly available on Hugging Face, allowing developers to download and run it independently. Z.ai confirmed that all traffic generated during the Ox Alpha test period was served on Chinese chips.
The Anonymous Testing Strategy and the Market Reaction
The Ox Alpha episode was not Z.ai’s first foray into pseudonymous model releases. The company previously tested GLM-5 anonymously under the name Pony Alpha, suggesting a deliberate pattern of using blind evaluations to benchmark models in real conditions before attaching the company’s name to them.
The mystery generated considerable speculation. Some developers suspected a Chinese laboratory was behind Ox Alpha; others floated the idea that Microsoft might be involved. The guessing game spread to Chinese social media, where users nicknamed the unknown model the “Niu Lai Model,” a reference to a low-budget animated film about a calf that had become a meme in China.
Z.ai ended the speculation on Wednesday, confirming Ox Alpha’s identity. The market response was swift: Z.ai’s stock climbed more than 8% in Hong Kong trading on Thursday, according to Quartz.
The share price move sits atop an already striking run since the company listed in Hong Kong in January under the corporate name Knowledge Atlas Technology, priced at HK$116.20 per share. It began trading on 8 January and changed its English name to Z.ai in July. Shares are now trading at around HK$1,100, roughly nine times the IPO price.
Z.ai’s financial disclosures paint a picture of rapid revenue growth alongside heavy spending. Last year the company generated 724.3 million Chinese yuan, or about $108 million, in revenue, more than double its 2024 total. It spent about 3.2 billion yuan on research and development and posted a net loss of about 4.7 billion yuan. The company is scheduled to report results for the first half of this year on Monday, making the timing of the Ox Alpha reveal particularly well-placed.
Z.ai competes directly with DeepSeek, Moonshot AI (the company behind the Kimi chatbot) and Alibaba’s Qwen team, all of them racing on model performance and price while navigating the same restricted access to leading-edge chips. The SGLang-based inference engine optimisation, and the leaderboard position it helped secure, will likely feature prominently in Z.ai’s narrative when those half-year results land.