Chinese AI models that publish open weights, downloadable worldwide from Hugging Face or ModelScope. Verify the license of each specific model before deployment — "open weights" does not always mean a permissive license.
| Model | Provider | Capabilities | License |
|---|---|---|---|
| DeepSeek-V3 | DeepSeek | chat, reasoning, coding | DeepSeek License |
| Qwen3 | Alibaba Qwen | chat, reasoning, coding, long-context | Apache-2.0 / model-specific |
| Qwen-Code | Alibaba Qwen | coding CLI agent | Apache-2.0 |
| MiniCPM-V 4 | OpenBMB | vision-language, on-device | Apache-2.0 |
| HY-World 2.0 | Tencent Hunyuan | 3D world model, generation | verify with provider |
| Hy-MT2 | Tencent Hunyuan | machine translation (zh/en/fr) | verify with provider |
| HunyuanOCR | Tencent Hunyuan | OCR | verify with provider |
| Intern-S1 | InternLM (Shanghai AI Lab) | scientific multimodal | Apache-2.0 |
| Intern-S2-Preview | InternLM (Shanghai AI Lab) | vision-language, reasoning (preview) | Apache-2.0 |
| Qwen3.6 | Alibaba Qwen | chat, reasoning, coding | Apache-2.0 (verify) |
| VoxCPM | OpenBMB | TTS, multilingual speech synthesis | Apache-2.0 (verify) |
- vLLM / SGLang — high-throughput GPU inference servers.
- llama.cpp — quantized CPU/GPU inference for smaller models.
- ModelScope — mirror of weights with China-region download speed.
This collection is curated by the acam-curator agent from data/models.yaml.
Submit additions through CONTRIBUTING.md.