China Telecom AI Releases Agentic Model for Single-GPU Deployment
29B parameters, 4B active: China Telecom's Xing4.0 brings agent-class models to single-GPU production.
Why it matters
A major frontier lab (China Telecom AI) shipped a mixture-of-experts agentic model designed for on-prem enterprise deployment at extreme compute efficiency. This signals a shift in the model-scale race: not bigger, but agent-capable and cheap to run. Practitioners can now evaluate whether lightweight agent models meet their autonomy use cases without GPU fleets.
The key facts
14 to knowModel: Xing4.0-29B-A4B (29B total parameters, 4B activated)
Deployment target: single GPU in production
Vendor: China Telecom Artificial Intelligence Technology Co., Ltd.
Release date: September 25, 2026
Type: lightweight agentic model; mixture-of-experts architecture implied
Use case: enterprise-grade agentic capabilities at low compute threshold
No pricing, inference latency, or throughput benchmarks disclosed
No independent capability benchmarks or comparisons provided
Model: Xing4.0-29B-A4B (29B total params, 4B activated via MoE)
Target: single-GPU enterprise deployment at scale
Vendor: China Telecom AI Technology Co., Ltd.
Classification: lightweight agentic LLM
No pricing, inference latency, or performance benchmarks disclosed
No comparison to competing agentic models (Qwen-Agent, DeepSeek-R3-Agent, etc.) provided
Go to the source
EnterpriseAIhpcwire.com
Publisher excerpt: BEIJING, Sept. 25, 2026 — China Telecom Artificial Intelligence Technology Co., Ltd. (China Telecom AI) recently officially released Xing4.0-29B-A4B, its next-generation lightweight agentic large model. With 29 billion total parameters and just 4 billion activated parameters, the model provides…