The mystery model was Z.ai the whole time. Last week a model called Ox Alpha showed up on OpenRouter. No vendor. No price. No parameter count. Free, fast, and quietly matching frontier models on coding. Nobody knew whose it was. On the 26th, Z.ai put their hand up. It was GLM 5.3 Flash. 320 billion parameters. Only 18 billion running at a time. The knowledge of a huge model at the price of a small one. Here's the part most people missed. Six days of free anonymous access wasn't generosity. 23.2 trillion tokens went through it in under a week. Roughly double the next model on the platform. And Z.ai says it ran on Chinese chips the whole time. Everyone's code was the load test. If you set it up last week, it's already broken. That name is dead. One line fixes it. Then three ways to keep running it. Free through Cline. Through OpenRouter at 7.5c per million, until September 9. Or entirely on your own machine, where nothing you send it ever leaves. That last one is what I'd use for client work.
4.43% ERengagement of this clip — below the author's average (8.36%)
top 77%outperforms 23% of the author's clips
0%of the author's average views
