A newer version of this model is available: zai-org/GLM-5.3-Flash
A friendly jade ox with ivory and gold horns and a gold alpha forelock. Ox Alpha. Revealed: GLM-5.3-Flash by Z.ai. The stealth model that became open weights.

Ox Alpha was GLM-5.3-Flash all along: open weights under MIT, and you can run it from our GGUF files.

By Stealth Models.

StealthMark 1.8.10 results for Ox Alpha (GLM-5.3-Flash): benchmark score 48.1, General Intelligence 41.7, Visual Intelligence 59.4. Open weights, and strongest at composing a whole scene: 77.7 on Beach. Every StealthMark result of Ox Alpha (GLM-5.3-Flash): Beach 77.7, Selfie 76.6, Ground Truth 63.8, Raccoon 55.1, Mona Lisa 49.5, Foresight 47.0, Newton's Mirror 27.6, Latch 3.6 (partial run, 4 of 5 levels).

Compare it with four other open models

The case file. First seen 20 Aug 2026, revealed 26 Aug 2026, developer Z.ai, context 1M tokens, input Text, image, tool calls Yes, speed in our tests about 89 tok/s. 20 Aug 2026: Debuts as Ox Alpha, with no developer named. 26 Aug 2026: Z.ai releases GLM-5.3-Flash and reveals it as Ox Alpha, with open weights under the MIT license. 28 Oct 2026, from us: Our GGUF files go live on Hugging Face: Q5_K_XL and IQ2_XXS, ready for llama.cpp.

See how it was revealed

Facts as text
Fact Value
First seen 20 Aug 2026
Revealed 26 Aug 2026
Developer Z.ai
Context 1M tokens
Input Text, image
Tool calls Yes
Speed in our tests about 89 tok/s
Access Active as GLM-5.3-Flash

Run it

Run it with our GGUF files: Q5_K_XL for near full quality, IQ2_XXS for 128 GB machines. Official weights: zai-org/GLM-5.3-Flash, MIT license.


StealthMark leaderboard
Text and images: CC BY 4.0, stealthmodels.com.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Space using stealth-model/ox-alpha 1