# Ox Alpha Is a New GLM Model. GLM-5.3 Flash Is Still My Guess

> Ox Alpha GLM is a new Z.ai GLM iteration. I still guess GLM-5.3 Flash because this stealth model takes images and video, unlike GLM 5.2 in my tests.

- Author: AI Kai
- Published: 2026-08-26
- Updated: 2026-08-26
- Topic: AI Dev Tools
- Tags: ox-alpha, glm, zai, openrouter, vision
- Canonical: https://hqman.me/blog/ox-alpha-glm/

Ox Alpha GLM is a new GLM iteration from Z.ai. My guess is GLM-5.3 Flash with vision, because GLM 5.2 in my tests could not see images.

Z.ai confirmed the family. The exact SKU is still unpublished.

## What is confirmed

Z.ai said Ox Alpha is a new iteration of its GLM series.

OpenRouter listed it as `stealth/ox-alpha`, with a 1M token context window plus text, image, and video input. OpenCode called it multimodal, with zero data retention.

I ran Ox Alpha on my 3D globe frontend test. The writeup is here: [Ox Alpha rebuilt my 3D globe dashboard](https://hqman.me/blog/ox-alpha-globe-test/).

The confirmed layer stops at GLM.

## Why I suspect GLM-5.3 Flash

I think Ox Alpha is GLM-5.3 Flash with vision. That is my read of the product, not an official model ID.

Community fingerprint posts pointed at Z.ai through tokenizer behavior and API error stacks. Those traces pointed at a new GLM drop. They did not prove a Flash SKU.

I keep GLM-5.3 Flash as a working label. A GLM with vision would explain the stealth launch and the traffic spike. I will drop that label if Z.ai publishes a different name.

## Why vision matters

My [GLM 5.2 vs Opus 4.8 frontend test](https://hqman.me/blog/glm-52-vs-opus-48-frontend-test/) hit a hard limit. GLM 5.2 could not see images. I had to describe the UI in text.

Ox Alpha takes text, images, and video.

I gave Ox Alpha a screenshot in OpenCode and it rebuilt a Three.js globe dashboard. GLM 5.2 could not take that screenshot as input. Coding quality can look similar across GLM versions. Image input cannot.

I will not map Ox Alpha to a named GLM SKU until the vision surface is in the model card.

## How people are using it

OpenCode said Ox Alpha processed 42T tokens in 5 days. OpenCode ranked it as the most used model after DeepSeek Flash's 56 day run.

OpenRouter said Ox Alpha is on track for close to 6 trillion tokens in a single day. The `ori` path they posted is:

```sh
ori [your favorite harness] --model stealth/ox-alpha
```

Use the [live Ox Alpha tracking thread](https://x.com/OpenRouter/status/2091912038264275103) for the current chart, then the [ori harness note](https://x.com/OpenRouter/status/2091912049999970543) if you want the same command in other coding agents.

## Caveat

Until Z.ai names the SKU, GLM-5.3 Flash stays a guess. Price and license claims stay out until I can verify them.

## Sources

- [OpenRouter: Ox Alpha token volume](https://x.com/OpenRouter/status/2091912024922177562)
- [OpenRouter: live Ox Alpha tracking](https://x.com/OpenRouter/status/2091912038264275103)
- [OpenRouter: ori harness](https://x.com/OpenRouter/status/2091912049999970543)
- [Ox Alpha 3D globe test](https://hqman.me/blog/ox-alpha-globe-test/)
- [GLM 5.2 vs Opus 4.8 frontend test](https://hqman.me/blog/glm-52-vs-opus-48-frontend-test/)