China Telecom Open-Sources Xing4.0, Runs on Single RTX 3090: what are the key facts?

China Telecom has open-sourced Xing4.0, a 29B-parameter Mixture-of-Experts (MoE) large model designed for code intelligence agents.

China Telecom Open-Sources Xing4.0, Runs on Single RTX 3090: what are the key facts?

Trained on Huawei Ascend 910C chips, the model achieves full-stack domestication from hardware to framework.

China Telecom Open-Sources Xing4.0, Runs on Single RTX 3090: what are the key facts?

After 4-bit quantization, memory usage drops to 15GB, enabling local execution on an RTX 3090 GPU.

China Telecom Open-Sources Xing4.0, Runs on Single RTX 3090: what are the key facts?

The model natively supports a 256K context window, offering capabilities for long-document processing and complex agent tasks.

China Telecom Open-Sources Xing4.0, Runs on Single RTX 3090: what are the key facts?

Xing4.0 is now available on GitHub and Hugging Face, where it has reached fourth place in trending repositories.

Physix Frontier · News Briefing Card (enbrief · Sep 21, 2026)

China Telecom Open-Sources Xing4.0, Runs on Single RTX 3090

KEY FACTS

China Telecom has open-sourced Xing4.0, a 29B-parameter Mixture-of-Experts (MoE) large model designed for code intelligence agents.
Trained on Huawei Ascend 910C chips, the model achieves full-stack domestication from hardware to framework.
After 4-bit quantization, memory usage drops to 15GB, enabling local execution on an RTX 3090 GPU.
The model natively supports a 256K context window, offering capabilities for long-document processing and complex agent tasks.
Xing4.0 is now available on GitHub and Hugging Face, where it has reached fourth place in trending repositories.

KEY DATA

29 billionTotal Parameters

~4 billionActive Parameters

~15 GBQuantized VRAM Usage

~96%Training Throughput Increase

PHYSIX OBSERVATION

China Telecom’s move precisely targets the pain points of data-sensitive enterprises by lowering the barrier to private deployment with consumer-grade GPU requirements. This full-stack domestic adaptation is not just a showcase of technological autonomy; it signals that the indigenous IT market is finally receiving a viable, lightweight AI office solution, likely accelerating AI adoption across government and enterprise sectors.

Source: enbrief original report ↗