A mystery model called Ox Alpha turns out to belong to Z.ai, and the company plans to give away its weights.
Z.ai, the Chinese lab behind the GLM model series, confirmed this week that Ox Alpha - a model that surfaced without any announcement and quickly drew comparisons to DeepSeek - is one of its own. The company says Ox Alpha is a new entry in the GLM lineup rather than a standalone experiment. Z.ai also confirmed it will release the model's weights rather than keep Ox Alpha closed. Beyond confirming its origin, Z.ai has not disclosed parameter counts, training details, or benchmark scores for the model.
The DeepSeek comparisons matter because DeepSeek's earlier releases proved a Chinese lab could match Western frontier labs on performance while giving weights away for free, upending assumptions about who controls the cutting edge. If Ox Alpha lives up to that comparison, it adds another data point to that trend. But so far the comparison rests on reputation and rumor, not published numbers.
Stealth releases make for good forum threads. They make for bad due diligence - wait for the eval scores before anyone calls this the next DeepSeek moment.