The Powerful Stealth AI Model ‘Ox Alpha’ Is Now GLM-5.3-Flash, and You Can Use It

Chinese AI company Z.ai has confirmed that the anonymous 'Ox Alpha' model that topped OpenRouter's popularity charts last week is its new GLM-5.3-Flash, an open-weight model now available across official platforms and on Hugging Face. Z.ai's stock price soared following the confirmation, and the model, which runs on Chinese-made AI chips, offers enhanced agentic capabilities and costs from $18 per month for coding plans, with weights exceeding 300GB for self-hosting.

The anonymous Ox Alpha AI model https://openrouter.ai/stealth/ox-alpha providers that popped up on OpenRouter, the unified AI platform, last week has been claimed by the Chinese company Z.ai http://z.ai . Now officially known as GLM-5.3-Flash https://z.ai/blog/glm-5.3-flash , the model is both powerful and cheap, challenging the capabilities and cost of popular models from Anthropic and OpenAI. When it was first teased, Ox Alpha became the most popular model for the week on OpenRouter and all of its traffic was served through Chinese-made AI chips. After it was confirmed to be the company behind the new model, Z.ai’s stock price soared, Bloomberg reports https://www.bloomberg.com/news/articles/2026-08-26/china-s-z-ai-made-ox-alpha-stealth-model-that-rivals-deepseek . Ox Alpha was described as a “frontier model built for efficient coding, sustained agentic work, and real-world production use,” according to an OpenRouter post on X https://x.com/OpenRouter/status/2090544970923184269 . In a follow-up post, the platform confirmed that the provider wouldn’t train on your prompts https://x.com/OpenRouter/status/2090544983141142722 . Z.ai’s GLM family are open-weight models https://hai.stanford.edu/ai-definitions/what-is-an-open-weight-model , where core components are publicly released, allowing anyone to download and use them. That’s different from open-source AI models, which make all pieces of the model available, including the source code, weights, and training data. Weights for GLM-5.3-Flash are available on Hugging Face https://huggingface.co/zai-org/GLM-5.3-Flash , another unified AI platform. Being open-weight puts the GLM family in contrast to closed-weight models from big US developers like OpenAI and Anthropic. Open-weight models, which are popular among Chinese companies, have been the subject of heated debate https://www.cnet.com/tech/services-and-software/open-source-ai-model-ban-proposal-cybersecurity-risks-news/ in the US because of the potential that users will change or misuse them. But big American companies, including OpenAI, have come out in support of open ecosystems. Meta recently said it would release the weights https://www.cnet.com/tech/services-and-software/meta-ai-glimmer-open-weights-news-2026/ of a previously closed model, Muse Spark 1.2. Like other recent AI models, GLM-5.3-Flash has enhanced agentic capabilities, such as using a browser and a computer, allowing the model to navigate web pages and open and interact with desktop apps. How can you access the GLM-5.3 Flash? In an announcement post on X https://x.com/Zai org/status/2092616204787626030 , Z.ai said the new model is available across all official platforms. The easiest and most direct way to access the model is to subscribe to one of Z.ai’s GLM coding plans, which start at $18 a month. The model will also provide 3x the usable quota of GLM-5.3. Because GLM-5.3-Flash is open-weight, you can also download it and use it on your own hardware — no paying for subscription or API costs. Although it might require some hefty computing power to do so, as the model weights themselves tip the scales at more than 300GB.