[REVISION HISTORY]

Z.ai confirms development of Ox Alpha AI model

Updated 6 times since CLSTR started tracking revisions of this situation.

What changed

2026-08-27 01:13 UTC → 2026-08-28 08:25 UTC · added ~~removed~~

The mystery surrounding the Ox Alpha stealth AI model has been resolved. The Chinese AI laboratory Z.ai, also known as Zhipu, confirmed that it developed the model as a stealth preview for its new GLM-5.3-Flash series. Initially appearing anonymously on the OpenRouter and OpenCode platforms, Ox Alpha gained rapid attention for its high performance in coding, reasoning, and long-horizon agentic tasks. The model features a 1,048,576-token context window and supports multimodal inputs including text, images, and video, though it rejects audio requests. During its limited free preview, the model saw massive usage, with reports indicating it processed 11.6 trillion tokens in just three days, supported by a claimed daily processing capacity of up to 100 trillion tokens. Following the confirmation of its origins, Z.ai released the model weights for GLM-5.3-Flash on Hugging Face. This move allows developers to build upon the architecture and highlights the increasing competition in the global AI landscape, as Chinese firms deploy highly capable, open-weight models at competitive price points. The model’s performance, which early testing suggested could rival or exceed established frontier models, was previously described by Stripe CEO Patrick Collison as “very impressive.” Recent reports indicate that the GLM-5.3-Flash model is specifically optimized to run on Chinese AI chips.

Versions

Only revisions since CLSTR began indexing content versions appear here. Select a version to see what changed compared to the one before it.