China Telecom releases Xing4.0-29B-A4B, a full-stack domestic lightweight AI agent model
中国电信开源首个全栈国产轻量级智能体大模型 Xing4.0-29B-A4B
This is the first full-stack domestic lightweight AI agent model from a major Chinese telecom operator, offering a viable open-source alternative to models like Qwen.
China Telecom released the Xing4.0-29B-A4B model on September 17th.
It has 29B total parameters and 4B activated parameters, supporting 256K context that can be expanded to 512K.
It scored 93.52 in the SuperCLUE benchmark, ranking third in the agent category, just 1 point behind two leading Qwen models.
After 4-bit quantization, its memory usage drops to 15GB, allowing it to run on consumer-grade RTX 3090/4090 GPUs.
The model is fully open-source, compatible with LLaMA-Factory, MindFormers, SGLang, vLLM, and KTransformers, and has been optimized for OpenCode, Claude Code, OpenClaw, and Hermes agent frameworks.
中国电信开源首个全栈国产轻量级智能体大模型 Xing4.0-29B-A4B
IT之家 9 月 19 日消息，中国电信于 9 月 17 日发布新一代星辰大模型 Xing4.0-29B-A4B，该模型总参数量 29B，激活参数仅 4B，原生支持 256K 上下文，可扩展至 512K， 是国内首个基于国产算力与国产框架完成训练、面向复杂工程任务深度优化的百亿参数大模型 。 官方表示，从训练芯片到推理部署， 全栈国产 —— 昇腾训练、国产框架适配、自研架构、开源生态，一条路走通。 从 SuperCLUE 评测结果来看，在智能体能力方面，Xing4.0-29B-A4B 得分 93.52，位列第 3 名，与两款头部 Qwen 模型的差距均不足 1 分。 据介绍，Xing4.0-29B-A4B 通过 4-bit 量化后显存占用降低至 15GB（比 FP16 省约 75%），RTX 3090、RTX 4090 这类 24G 消费级显卡就能本地跑长上下文任务。 Xing4.0-29B-A4B 延续星辰大模型的开源路线，对主流开源生态做了全面兼容：支持 LLaMA-Factory、MindFormers 等微调架构；适配 SGLang、vLLM、KTransformers 等主流推理框架；Agent 框架对 OpenCode、Claude Code、OpenClaw、Hermes 做了定向适配调优，深度对齐各框架格式规范。 星辰大模型 Xing4.0-29B-A4B 已上线开源社区，IT之家附相关链接： HuggingFace： https://huggingface.co/XingChen-AGI/Xing4.0-29B-A4B GitHub： https://github.com/XingChen-AGI/Xing4.0-29B-A4B 魔搭： https://modelscope.cn/models/XingChen-AGI/Xing4.0-29B-A4B Gitee： https://gitee.com/xingchen-agi/xing4.0-29b-a4b 魔乐： https://modelers.cn/models/XingChen-AGI/Xing4.0-29B-A4B
This report is categorized under AITOP's "Models" desk, focusing on China's AI industry and domestic foundation model ecosystem. Within our cross-regional analytical framework, it directly maps to Chinese AI companies expanding and localizing across Southeast Asia and US-China foundation model benchmarks and commercialization comparison and AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include Xing4.0-29B-A4B, Qwen, SuperCLUE, open-source models.
Source attribution: curated from IT之家. Cross-source corroboration: 2 independent sources have covered this development (including 量子位), confirming industry-wide signal strength.
- 量子位09-20 12:22Source
