Published on •Updated
A strange new AI system called Ox Alpha has created a storm of confusion among tech analysts and users alike.
Nobody knows who made this model, which dropped on OpenRouter as an anonymous proprietary stealth model on 20 August.
While it is not open source, it is currently offering a free preview until 27 August, after which it is yet to be determined whether the model will be chargeable or not.
Although the maker and owner of this model is currently unknown, there are suspicions of it potentially belonging to China-based Zhipu AI’s unreleased GLM-5.x series. However, these speculations have not been confirmed yet.
This has led to increased concerns about China potentially overtaking Western powers like the US in the AI race by consistently rolling out highly sophisticated and often unclaimed models.
In the past other Chinese companies such as Alibaba and ByteDance have released models without immediately taking credit for them.
This can be to bypass international export scrutiny, avoid immediate regulatory blowback and test high-capacity models without tying liability to a specific corporate entity.
Why Ox Alpha is so powerful
Coding benchmarks such as DeepSWE suggest Ox Alpha is as good as, if not better than, several top models from industry giants including OpenAI and Anthropic.
The model already boasts a token context window exceeding 1 million, and can process images, text and video content with almost unlimited usage, despite its power and current free tier.
It also has a significant output capacity of up to 131,072 tokens in a single response, allowing it to write expansive, detailed scripts.
Ox Alpha is designed for long-horizon tasks such as complex programming, production-scale data processing and sustained agentic workflows.
Its website currently states that prompts are “retained by the provider” but not used for training, though this has not been independently verified.
The model highlights the scale of China’s AI ambitions, with the state treating artificial intelligence as a national and strategic priority. This allows Chinese companies to cut costs, roll out models quickly, and pivot fast when needed.
US AI companies, by contrast, often face delayed or staggered rollouts, hampered by expensive proprietary models and fast-changing export controls and chip bans.