Startup Z.AI confirmed to be behind the mysterious model

In an August 26 blog post, Z.AI said Ox Alpha was actually an anonymous preview of the company's new GLM-5.3-Flash model, tested on the OpenRouter and OpenCode platforms to gather user feedback before release.

The startup also announced that it has made the model's weights public on the Hugging Face platform, allowing programmers to download, edit and develop products based on it. This is a multimodal model, meaning it can work with text, images, and videos, and has a context window of up to one million tokens (the basic unit of input data that the AI ​​model processes to create output).

Ox Alpha appeared last weekend on OpenRouter - an API gateway platform that helps connect applications with hundreds of different AI models through a single API key. This AI is provided as a "secret model" from an anonymous third party.

According to the description, Ox Alpha is "a reasoning model designed for programming, suitable for long-term software engineering areas, complex reasoning, and workflows that combine text with visual context". The model provides the ability to process up to 100,000 billion tokens per day, about 100 times the number of AI tokens that multinational payment card service corporation Visa uses in a month.

However, according to Business Insider, the special feature is that Ox Alpha allows free use for one week with "almost unlimited" rights. This attracted great attention in the context of many companies, including Tesla and Microsoft, limiting tokens to save costs.

On Twitter, many people expressed excitement about Ox Alpha. Patrick Collison, CEO of technology company Stripe, tried the model and commented "very impressive".

Technology news site Wccftech previously speculated that this AI could be the product of a Chinese artificial intelligence laboratory and that Z.AI is "one of the possibilities". The Beijing-based startup also tested an anonymous GLM-5 model under the name Pony Alpha.

TechCrunch commented that Ox Alpha's appearance increases the possibility of cheap, effective Chinese models taking significant market share from expensive American advanced model development companies such as OpenAI and Anthropic.