China Telecom Open-Sources Xing4.0-29B-A4B Agentic MoE Trained End-to-End on Ascend, Pandaily reports
Reported by Pandaily · not yet confirmed by the company or a second independent outlet. We update this page when it is.
China Telecom AI released Xing4.0-29B-A4B, a 29B-parameter Mixture-of-Experts model with 4B active parameters and 256K context, trained end-to-end on Ascend NPUs with MindSpore and published with open weights on Hugging Face and ModelScope, Pandaily reported.
China context
- Original name
- 星辰4.0-29B-A4B
- Outside China
- Open weights · pandaily.com
- Claims
- Company-reported; not yet independently evaluated
- For builders
- Developers outside China can download the weights from Hugging Face and run the model on compatible hardware, though Ascend NPU availability may limit local inference.
- For investors
- The release signals China Telecom's commitment to domestic AI infrastructure and could pressure other Chinese telecoms to follow suit.
China Telecom AI released Xing4.0-29B-A4B, a 29B-parameter Mixture-of-Experts model with 4B active parameters and 256K context, trained end-to-end on Ascend NPUs with MindSpore and published with open weights on Hugging Face and ModelScope.
The model uses a Mixture-of-Experts architecture with 29B total parameters and 4B active parameters, supporting a 256K context window. It was trained end-to-end on Ascend NPUs using the MindSpore framework, demonstrating a fully domestic Chinese AI training stack.
Chinese AI developers gain access to a high-capacity open-weights model that runs on domestic hardware, reducing reliance on Nvidia GPUs and potentially lowering inference costs for agentic applications.
Enterprises can deploy an open-weights agentic model on Ascend hardware without export-controlled GPUs, potentially reducing infrastructure costs and compliance risks in China.
Observable next signals include the model's download count on Hugging Face and ModelScope, community fine-tunes or benchmarks, and whether China Telecom releases larger or more specialized variants.