So literally the one everyone guessed.
Introducing GLM-5.3-Flash
- Leading capabilities at a highly competitive price
- Natively multimodal with a 1M-token context window
- A 320B-A18B model released under the MIT License
- Previously previewed as Ox Alpha, running entirely on Chinese AI chips
Blog: z.ai/blog/glm-5.3-f…
Available now across all official platforms:
Weights: huggingface.co/zai-org/GLM-5.…
API: docs.z.ai/guides/llm/glm…
Coding Plan: z.ai/subscribe
ZCode: zcode.z.ai/en
Chat: chat.z.ai
AutoClaw: autoclaw.z.ai
Aug 26, 2026 · 3:39 PM UTC
10
3
293
19,881
@neetcode1 It is running on Chinese chips is the surprising part, at least for me
0
0
5
497
@neetcode1 "Everyone is wrong about ox-alpha, and its going to be a HUGE surprise", i dont get these people honestly
0
0
2
479
@neetcode1 Well the surprise was on an other level that no one guessed. Running on only Chinese chips 😄
0
0
0
236
@neetcode1 Hey @neetcode1 the Neetcode website seems to be down for me. I can open the roadmap or any of the videos or questions. Just reporting the outage.
0
0
0
12
@neetcode1 @grok can you help me get the context behind this post. Also what’s ox-alpha
1
0
0
27
@neetcode1 is this what you are reduced to? Commenting about AI model releases?? What a sad world...
0
0
0
54
@neetcode1 Yep, but the scale behind it is even more wild - 42 trillion tokens in 6 days with those benchmarks. We actually broke this down here: x.com/AGTPinsights/s…
Ox Alpha just got unmasked today. Here's what you need to know.
OpenCode said Ox Alpha, the free stealth model, hit 42 trillion tokens processed in just 6 days, making it the most-used model on the platform after DeepSeek Flash's 56-day run. A full reveal was teased for later in n confirmed it: Ox Alpha is a new GLM model from China's Z.AI, also known as Zhipu, widely believed to be GLM-5.3 Flash. Z.AI told Bloomberg it will release the model's weights tonight.
Z.AI's GLM-5.3 launched on August 14 with a 1,048,576-token context window, scoring 66.9 on DeepSWE v1.1 and 28.3 on Terminal Bench 3.0. Speculation is swirling that Nvidia backed the compute behind roughly 100 trillion free tokens as an open-source marketing push. Independent testing from Vectal Labs put Ox Alpha at 84.3 on GPQA Diamond, versus GLM-5.3's reported 91.7, though they only managed 2 runs due to slow inference.
Key numbers:
- 42T tokens processed in 6 days
- ~100T free tokens reportedly given away
- 66.9 DeepSWE v1.1 score for GLM-5.3
- 84.3 vs 91.7 on GPQA Diamond, Ox Alpha vs GLM-5.3
- 1,048,576-token context window
Weights are set to drop tonight, closing out days of stealth-model guessing.
0
0
0
300