Ox Alpha arrived with the kind of combination that makes AI developers stop what they are doing: strong coding claims, broad access and no named creator. Its performance is drawing attention; its provenance is drawing a detective hunt.

The model surfaced on OpenRouter on Thursday as an anonymous “stealth model,” pitched for coding, sustained agentic work and production use. OpenCode said the preview would be free for a week with “near unlimited usage,” backed by claimed capacity of 100 trillion tokens a day.¹ Stripe chief executive Patrick Collison tested it and called it “very impressive.”¹

That early excitement quickly turned into an attribution contest. One prominent theory placed Ox Alpha at Z.ai, the Chinese lab associated with GLM models. The case rested on reported similarities in tokenizer behaviour and responses, as well as Z.ai’s earlier anonymous testing of GLM-5 under the name “Pony Alpha.” But the same reporting stressed that the evidence was far from settled: a competing analysis pointed instead toward Microsoft’s MAI family.¹

By Friday and Saturday, the uncertainty had become the story. AI analyst Andrew Curran said GLM had initially been the leading guess, but that “people seem less sure of anything” by the following morning.² Online debate reflected that split, with some users insisting the model could not be Chinese and others professing high confidence that it was.²

The search has since spilled beyond benchmark clues and corporate genealogy. A separate account of the mystery points to China’s eccentric viral-film culture as a possible source of hints, underscoring how little hard evidence has emerged about the model’s operator.³ For now, Ox Alpha’s most convincing feature may be its output — and its most effective marketing may be the unanswered question of who made it.