Telecom's Xing4.0-29B-Intent open-sourced: intent + slot filling, beats Jev by 12.6 points zero-shot
智东西 · wechat · 2026-10-19
China Telecom's XingChen AGI lab open-sourced Xing4.0-29B-Intent, extending intent understanding beyond Jev-style fast decisions to joint intent classification and parameter extraction, outputting unified category+args JSON with sub-150ms latency. Key facts:
- Zero-shot Micro-F1 of 70.27, 12.62 points above Jev's best config and above GPT-4.1; strict all-correct score of 61 vs Jev's 21–48
- Overall score 58 across 48 public datasets, 9 points ahead of runner-up Gemini-3.6-Flash
- Built on the USP (Unified Schema-driven Semantic Parsing) framework unifying NER, relation and event extraction and spoken language understanding
- Ships 2.75M SFT samples (rebuilt from 61 public datasets, 19 domains, 52 languages), inference SDK, Apache 2.0
Zhidongxi's 11 real-scenario tests (multi-device smart-home control, dual work orders, multi-turn meeting edits preserving unchanged fields, ignoring instructions embedded in memo text) show solid potential for smart cockpits, enterprise workflows and personal agents.
Related event: China Telecom Open-Sources Xing4.0-29B-Intent for Intent Understanding(3 posts)→
More from coding & agent
- Vercel adds no-account deploys: drag and drop a file to get a live URL — steren · 2026-10-23
- Vercel ships 'vercel deploy --temporary' for ephemeral agent-built deployments — cramforce · 2026-10-23
- Running multiple agents out of one Notion workspace as a shared task board — nbaschez · 2026-10-23
- We tested 12 LLM classifiers — the new classifier hype is LARP — KlausCodes · 2026-10-23
- Dev builds ESP32 desk display showing Claude Code and Codex agent status via local Go helper — tonylook · 2026-10-23
- Let your agent read its own traces: AgentsView + history tool debugging workflow — dbreunig · 2026-10-23