Today for AI
HOT RADAR
llmHEAT 11.0°

AI CURATED HOT STORY · 10/22/2026

China Telecom Open-Sources Xing4.0-29B Intent Model for Structured Parameter Extraction

1 reports archived1 independent sourcesupdated 10/22/2026, 15:11:14
Synthesis & Latest Updates
1 Sources Cross-Validated

China Telecom has released and open-sourced the specialized intent understanding model Xing4.0-29B-Intent, along with 2.75 million training samples and inference images. Utilizing a Unified Schema-driven Semantic Parsing (USP) framework, the model handles both intent classification and parameter extraction, outputting standardized JSON. Compared to general LLMs, it offers optimized low latency, lower cost, and higher reliability for vertical scenarios like customer service routing and smart home control.

LATEST/China Telecom has released and open-sourced the specialized intent understanding model Xing4.0-29B-Intent, along with 2.75 million training samples and inference images. Utilizing a Unified Schema-driven Semantic Parsing (USP) framework, the model handles both intent classification and parameter extraction, outputting standardized JSON. Compared to general LLMs, it offers optimized low latency, lower cost, and higher reliability for vertical scenarios like customer service routing and smart home control.

HEAT TRENDHourly heat curve

2 fully observed hours
Fewer than 3 fully observed hours; trend not plotted yet.

TIMELINECoverage timeline

Total 1 reports · Latest first
  1. 量子位官网T2·68 pts
    • Uses the USP framework to unify intent recognition and parameter extraction into a 'Category + Parameter' hierarchy, directly outputting JSON.
    • Optimized for high-frequency calls with average outputs of dozens of tokens and sub-100ms latency, significantly reducing decision costs.
    • Already deployed in China Telecom's customer service and smart home speakers, validating its effectiveness in dialogue intent recognition and ticket routing.
China Telecom Open-Sources Xing4.0-29B Intent Model for Structured Parameter Extraction | Today for AI