Open-weight Chinese LLMs
As of 2026-11-04.
| Model family | Example open release | Documented license | Source |
|---|---|---|---|
| DeepSeek | DeepSeek-V4-Pro / Flash | MIT | HF model card |
| Qwen | Qwen3.8-27B | Apache 2.0 | LICENSE |
| Qwen Max-class | Larger MoE / Max checkpoints | Often separate Qwen license | Check each repo |
| GLM | Varies by generation | Release-specific | HF org + card |
Open weights ≠ free hosted API. GPU, safety filters, and export/compliance rules are your responsibility.
FAQ
Can I use DeepSeek V4 weights commercially?
Hugging Face model cards for DeepSeek-V4-Pro and DeepSeek-V4-Flash state MIT licensing for repository and weights. Keep the copyright notice; still run your own compliance review.
Is every Qwen model Apache 2.0?
No. Mid-size releases such as Qwen3.8-27B are documented as Apache 2.0; some Max/flagship checkpoints use a separate Qwen license with additional conditions. Read the LICENSE in the specific repo.