Xing4.0-29B-Intent Ships Unified JSON Extraction

The compelling angle is packaging schema-guided extraction as a deployable developer primitive, not merely releasing another general-purpose LLM.

• –Unifies intent classification, slot filling, NER, relation extraction, and event extraction under one `category + args` format.
• –The 4B-active MoE design could make specialized extraction more practical to self-host than a dense 29B model.
• –Its reported 81 intent-understanding score is promising, but the overall score drops to 58, signaling that real-world schema validation remains essential.
• –The dataset, reusable schemas, and SDK reduce the work required to adapt the model for assistants, routing, and enterprise workflows.
• –Apache-2.0 licensing makes it viable for commercial deployments that need local or private inference.