China Telecom AI has released Xing4.0-29B-A4B, a lightweight open model for running complex autonomous workflows on a single desktop GPU.
Built with 29 billion total parameters, the architecture activates four billion parameters per token during inference. That structural split allows the software to balance low-latency responses with sustained multi-step task execution, lowering hardware barriers for organisations deploying automated software engineering and analysis pipelines.
A spokesperson for China Telecom AI said: “We believe the future of AI lies not in ever-larger models, but in making powerful intelligence accessible to everyone, affordable, and deployable anywhere. Xing4.0-29B-A4B proves that a compact model can also deliver agentic capabilities on par with far larger models—whether on a developer’s desktop, or in an enterprise production environment.”
Mixture of experts across 256K context
Engineers configured the system on a mixture-of-experts design that pairs sparse activation with a 256K-token context window. This large capacity lets the model ingest entire software repositories, review extensive technical documentation, and trace multi-file dependencies.
Instead of functioning strictly as an interactive conversational interface, the system parses multi-part operational instructions, issues calls to external software tools, and generates structured code outputs.
Memory efficiency forms the backbone of the release. The company applied low-bit quantisation alongside custom memory allocation methods to compress the full runtime footprint into 15 GB of graphic memory.
Individual technical staff can test and run local automated workflows – including script execution, repository inspection, and data extraction – without transmitting proprietary internal data outside their local hardware environments. Enterprise data centres can similarly bypass high-density compute clusters to serve automated agent tasks from modest hardware nodes.
On standard technical evaluations, the system recorded a score of 75 out of 100 on the SWE-bench Verified benchmark. That evaluation tests an engine’s proficiency at inspecting multi-file software repositories, identifying technical bugs, generating correct patches, and verifying code execution against real issue tickets. The release also verified high compliance rates on structured JSON output generation, function-calling precision, and long-sequence retrieval tests.
China Telecom verifies its AI in live production environments
China Telecom’s operations teams integrated Xing4.0-29B-A4B into the group-level customer service platform. Within that environment, the model routes customer inquiries, plans multi-step remediation procedures, and triggers external resolution systems, raising both first-contact resolution rates and general case handling speed.
The company also deployed the software into interactive home service environments, providing real-time technical assistance across mid-screen domestic hardware.
Model weights and code dependencies are distributed through the XingChen-AGI repositories on GitHub and Hugging Face under an open-ecosystem distribution model. The software integrates directly with mainstream open-source training, serving, and agent development environments.
China Telecom’s engineers confirmed compatibility across standard commercial silicon as well as domestic Chinese accelerator platforms. The division currently maintains more than 500 operational enterprise workflows across 20 commercial sectors through its Xingchen model family, and plans to introduce larger-parameter models within the Xing4.0 family during the coming months.
See also: Nokia and Microsoft automate telecom networks with AI agents
Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events including the IoT Tech Expo and Cyber Security & Cloud Expo. Click here for more information.
Telecoms is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.
