GLM 5.5 leak: China's next AI could fit on your home PC

China's Zhipu AI, also known as Z.ai, appears to be finishing the next version of its flagship model. Today a screenshot of a single line of code spread on X, describing what one of the company's teams is working on: a "next-generation GLM base model (GLM-5.5)", end-to-end post-training for coding, quant-finance training environments and a dedicated coding sandbox. The post with the image passed 25,000 views within a few hours.

The juiciest part is in the same line: internally the company is post-training three models at once, with 32, 122 and 256 billion parameters, and then merging them into a shared base. If the smallest one ships with open weights, as Zhipu did with earlier GLM models, it would take roughly 16 GB at 4-bit quantization and fit on a single gaming GPU. The line also mentions a separate research topic on "self-evolution" in financial coding, meaning a model that learns to improve itself.

The author of the post doesn't say where the screenshot came from. Zhipu hasn't confirmed anything or given a date. The previous GLM-5.3 came out more than two months ago, and we already covered what it can do. People in the replies are betting on a November release, but that's just guessing.

For now it's one line of code and a lot of hope. A very specific line, though.