# Z.ai Confirms Open Weights for Ox Alpha GLM

> Published 2026-08-26 · https://www.promptzone.com/theo_jung/zai-confirms-open-weights-for-ox-alpha-glm-5c35

Z.ai confirmed that **Ox Alpha** is a new GLM-series model and will release its weights openly. The announcement followed earlier reports of a stealth project positioned to rival **DeepSeek** models. The story first gained traction on [Hacker News](https://www.bloomberg.com/news/articles/2026-08-26/china-s-z-ai-made-ox-alpha-stealth-model-that-rivals-deepseek), where the thread reached 170 points and 68 comments.

## What Ox Alpha Is

Ox Alpha belongs to the GLM model family developed by Z.ai. It was built as a stealth project before public confirmation. The company stated it will publish the weights, shifting the model from closed testing to community use.

## How the Release Works

Z.ai plans to distribute the weights directly rather than through an API-only channel. No specific license details were shared in the initial confirmation. Early HN comments focused on whether the release will include training code or only inference weights.

## HN Community Reaction

The discussion highlighted three main points:
- Interest in how Ox Alpha compares on benchmarks to current DeepSeek releases
- Questions about the exact parameter count and training data sources
- Concerns over export controls affecting weight downloads outside China

## Comparison with DeepSeek Models

| Feature          | Ox Alpha (expected) | DeepSeek-V3     |
|------------------|---------------------|-----------------|
| Series           | GLM                 | DeepSeek        |
| Weight release   | Planned             | Already open    |
| Reported focus   | Rival performance   | Math and code   |
| Origin           | Z.ai (China)        | DeepSeek (China)|

DeepSeek models currently lead in open-weight math and coding tasks. Ox Alpha's release will test whether Z.ai can match that performance with a GLM backbone.

## Who Should Watch This Release

Researchers tracking Chinese open models should monitor the weights for direct comparison tests. Developers building on GLM architectures gain a new base model once released. Teams already committed to DeepSeek-V3 or Qwen2.5 can skip unless Ox Alpha shows clear gains on specific benchmarks.

## Practical Next Steps

Wait for the official weight upload on Hugging Face or Z.ai's repository. Run standard evaluation suites such as MMLU and HumanEval once available. Compare latency and accuracy against DeepSeek-33B and DeepSeek-67B on identical hardware.

> **Bottom line:** The confirmation moves Ox Alpha from rumor to scheduled open release, giving the community a direct competitor to test against current DeepSeek weights.

Z.ai's decision to publish weights adds another high-performing GLM variant to the open ecosystem and will likely trigger fresh benchmark comparisons within weeks.