China Telecom Open-Sources Xing4.0-29B-A4B, Its First Fully Domestic-Stack Lightweight Agent LLM
Summary
China Telecom released and open-sourced its new-generation Xing (Star) model Xing4.0-29B-A4B on September 17, with 29B total parameters, only 4B activated, native 256K context extendable to 512K. It is billed as China's first 10B-plus-parameter model trained entirely on a domestic stack — Ascend training chips, domestic frameworks, in-house architecture — and scored 93.52 on SuperCLUE's agent benchmark, third place and less than one point behind two leading Qwen models. After 4-bit quantization its VRAM footprint drops to 15GB, letting consumer GPUs like the RTX 3090/4090 run long-context tasks locally.
Bubble analysis
This is a model release: it carries no investment figure, valuation or compute purchase, so it offers almost no direct evidence on whether AI is in an investment bubble. Its only indirect relevance is that a fully domestic training stack and local deployment on consumer GPUs suggest model efficiency is improving and the compute bar per model is falling — which over the long run could soften demand for massive capex, but that chain is too long to count as bubble evidence either way.
Related signals
linked by shared tags
- Huawei Open-Sources Pretraining, SFT and Post-Training RL Code for openPangu-2.0#ascend#open-source#llm
- Ant Ling Releases Ling-3.1-flash Model: ~560B Total Parameters, 25B Activated Parameters#open-source#llm
- StepFun Releases Flagship Model Step 5 Preview: Enters Global Top Three Open-Source Models on AA Leaderboard, Open-Sourcing on October 15#open-source#llm
- Bilibili Open-Sources the Index-Translate Multilingual Translation Model Family, Text Models Cover 150 Languages#open-source
- Lao Wang on Industry: Compute, Cash Flow and Business Models — How the World's Leading LLM Companies Operate, and the US-China Gap#llm