Z.AI From China Confirms It Built Ox Alpha, and the Model Is Going Open Weight
Trending Topics Jakob Steinschaden ● Covered by 3 sources
Z.ai says the mystery model Ox Alpha is its new GLM release. It’s going open weight tonight, after a week of wild benchmark chatter.
Based on reporting by Trending Topics, Jakob Steinschaden — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
For a week, a nameless model had people in AI circles arguing over who built it. Now the mystery has a name: Z.ai, the Chinese company formerly known as Zhipu AI, told Bloomberg that the model appearing as Ox Alpha is a new version of its GLM family. The weights are due out tonight, which makes this an open weight release, not just another tease.
That fits Z.ai’s pattern. The company has been putting GLM models out with open weights on a regular basis, and those releases have kept it in the conversation internationally. The latest twist was a pause on GLM-5.3: Z.ai delayed that weight release by about two weeks because the model was apparently good at finding security vulnerabilities. During that stretch, access stayed limited to selected security partners.
Ox Alpha first showed up last Thursday with no announcement and no obvious vendor attached. On OpenRouter it appeared under the provider label “Stealth,” while OpenCode, Cline and a Nous Research portal all carried it too. The pitch was bold: free access, a one-million-token context window, text, image and video input, and tool calling. OpenCode said it could handle 100 trillion tokens a day; Nous Research said even more.
The first public tests only fueled the noise. On a small slice of the DeepSWE coding benchmark, Decrypt said the model hit 80 percent, ahead of Claude Fable 5 and GPT-5.6 Sol. On the full 113-task run, that advantage shrank to about 63 percent, roughly where GPT-5.6 Sol sits. Patrick Collison called it “very impressive” on X, and that helped push the guessing game into overdrive.
The speculation list was sprawling: Zhipu, yes, but also Microsoft’s MAI family, DeepSeek, Xiaomi’s MiMo and Google. Meanwhile, OpenRouter usage charts showed Ox Alpha climbing, helped by coding agents like Claude Code and Hermes Agent. If the official rankings end up matching what these community tests have suggested, Z.ai’s place among open model makers should get a lot harder to ignore.
My take — AI-written commentary, not fact-checked reporting
The open-weight crowd keeps proving the same annoying point: the serious action isn’t always behind closed doors. Z.ai looks like it understands that better than most Western labs, which is why the model race keeps getting flatter and more public. Closed models still lead, but the moat is starting to look more like a puddle.
Read more about this at: Trending Topics