Huawei · 2× · tolest 18. Sept. 2026
Pangu (or openPangu) is Huawei's family of large AI language models developed since July 2021, now available in version openPangu 2.0 as an openly accessible model. The current Pro variant is a Mixture-of-Experts model with approximately 505 billion total parameters (roughly 18 billion active per Token), a context window of 512,000 Token, and was trained entirely on Huawei's proprietary Ascend-910B-NPUs without Nvidia hardware. The models are provided via GitCode (Ascend-Tribe-Community), Hugging Face, and Huawei Cloud ModelArts Studio and are subject to a proprietary, non-free Huawei license (OpenPangu Model License Agreement). Official pricing per Token for API access via ModelArts Studio had not been published at the time of research.
Features
| Context Window (Tokens) | 512,000 tokens |
| License | OpenPangu Model License Agreement Version 2.0 (Huawei's own license, not a standard open-source license) |
| Multimodality | Pangu (base model family) is multimodal (text, image, video, etc.); openPangu-2.0-Pro itself is primarily a language model |
| Platform | Available via GitCode (Ascend Tribe community), Hugging Face, and Huawei Cloud ModelArts Studio |
| Release Date | openPangu-2.0-Pro released August 4, 2026 (original Pangu launched July 2021) |