The leading theory has been Z.ai, which once tested GLM-5 anonymously under a different name — though the same tokenizer clues have also been read as Microsoft's.
A reasoning model named Ox Alpha turned up on OpenRouter last Thursday with no developer attached to it, credited only to an outside supplier that goes unnamed, and free to use. Nobody has publicly said who built it.
OpenCode, a project that offers an open-source AI agent for writing code, said in a post on X last week that Ox Alpha would stay free for seven days, with usage all but uncapped, and that the provider behind it had capacity to serve 100 trillion tokens per day. That claimed daily figure is roughly 100 times the volume of AI tokens Visa has said it gets through in a full month. Among those trying the model is Patrick Collison, Stripe's chief executive, who wrote on X that he found it very impressive.
The hunt for a maker has run largely on tokenizer clues, and it has split. Early speculation pointed to a Chinese AI lab: developers noticed that Ox Alpha's tokenizer behavior and responses resembled the GLM family's, and the tech publication Wccftech suggested Z.ai — the lab behind GLM-5, which it had previously tested anonymously under the label Pony Alpha — as one possible creator. That reading arrives at a moment when Chinese labs are increasingly challenging their American rivals, with largely open-source models that cost far less while coming close to what the best American systems can do.
The same tokenizer evidence has since been read the other way. Wccftech has highlighted a competing analysis under which the clues point instead toward Microsoft's MAI family of models.
The evidence pointing to a maker is nowhere near conclusive. Andrew Curran, an AI analyst, posted on X on Saturday that the GLM theory had been the leading explanation on Friday night, though by Saturday morning people appeared less sure of any account. The model keeps circulating among developers all the same, and still nobody has claimed it.