China Telecom's XingChen lab open-sources Xing4.0-29B-Intent, a 29B MoE with 4B activated params
FellMentKE · x · 2026-10-26
China Telecom's XingChen AGI lab released and open-sourced Xing4.0-29B-Intent (Apache-2.0), a schema-guided model for intent recognition and structured information extraction.
- 29B MoE with only 4B activated parameters
- Combines intent classification with slot filling, plus entity, relation and event extraction — returning actionable parameters (room, brightness, color temperature), not just a label
- Ships with a companion dataset and XingIntentSDK; available on GitHub, Hugging Face and ModelScope
For agent builders, merging "understand the command" and "fill in execution params" into one specialized small model is a practical way to simplify agent pipelines.
More from coding & agent
- A visual guide to the 7 layers of scalable system design — goyalshaliniuk · 2026-10-26
- Solo dev renders 52-shot cinematic game trailer with Claude Code on one laptop for a few dollars — freechoice · 2026-10-26
- Open-source Buddy app runs Codex CLI on Android, builds and installs a game from one sentence — 3amae404 · 2026-10-26
- Killed the vector DB for SQLite BM25 (0.35ms) and wrote a 3.6ns Rust sentinel for agent loops — OpenComplaint524 · 2026-10-26
- Official A2A CLI simplifies testing Agent2Agent servers from the terminal — bibryam · 2026-10-26
- Claude Code rebuilt its own ComfyUI pipeline to fix prop continuity in a 2-min AI film — Odd_Definition_9724 · 2026-10-26