China Telecom AI model releases

China Telecom AI (TeleAI, the Institute of Artificial Intelligence of China Telecom) is the AI research division of state-owned carrier China Telecom, established in 2023 and led by academician Xuelong Li. It develops the XingChen (Starlight) foundation-model line, publishing open weights under the XingChen-AGI handle, and trains its models on the domestic Ascend NPU / MindSpore stack. First tracked model is Xing4.0-29B-A4B (Sep 2026). This page collects the lab's model releases, lifecycle events, source links, and model metadata in one crawlable record.

Models

Labs

Open

Recent

All models·Latest releases·Open-weight models·Changelog·Home

Most recent in this set

Sep 22, 2026

Xing4.0-29B-A4B

TeleAI

Sort

1 model

Xing4.0-29B-A4B

Available

China Telecom AIOpen source

Xing4.0-29B-A4B is China Telecom AI (TeleAI)'s open-weights agentic large language model, released 2026-09-22 under Apache 2.0 via the XingChen-AGI GitHub/Hugging Face handle. A Mixture-of-Experts model with 29B total parameters and ~4B active per token (40 layers, hidden size 3584) that deploys on a single GPU in ~15 GB. Purpose-built for the agentic era: multi-step planning, tool/function calling, long-document analysis and code, with an enable_thinking switch for extended reasoning. Natively supports 256K context, extensible to 512K. Reports SWE-bench Verified 75.0, top of its parameter class, and is deployed in China Telecom's customer-service platform. Trained end-to-end on the Ascend NPU platform with the MindSpore framework.

MoE29B262K ctxSep 22, 2026