China Telecom open-sources Xing4.0-29B-Intent for structured commands

China Telecom's XingChen AGI lab has open-sourced Xing4.0-29B-Intent, a model that classifies user intent and extracts parameters into a unified JSON structure in a single pass. The team reports sub-150 ms inference latency, a zero-shot Micro-F1 of 70.27 and an exact-match score of 61. Code is on GitHub under Apache 2.0, with weights on Hugging Face and ModelScope.

• Model combines intent classification and parameter extraction in one pass
• Inference latency under 150 ms, zero-shot Micro-F1 of 70.27
• Trained on 2.753 million samples from 61 datasets in 52 languages
• Code on GitHub under Apache 2.0, weights on Hugging Face and ModelScope