LlamaIndex adds calibrated confidence scores to LlamaParse Extract, field by field

llama_index · x · 2026-09-22

LlamaIndex shipped Grounded Confidence for LlamaParse Extract, extending the recent push around calibrated confidence scores for discrete decisions to general schema-guided document extraction. The scores cover primitive types (bool, int, float — unbounded values included) and free-form text, and each extracted value carries a bounding box back into the source document.

The practical use is human-review triage: set a threshold, auto-accept high-confidence fields, and route low-confidence ones to auditors. Available on the Cost Effective, Agentic, and Agentic Plus tiers. Jerry Liu frames it as essential when agents and workflows depend on extracted data.

Related event: LlamaParse Extract Adds Grounded Confidence Scores(2 posts)→

Original post →

More from coding & agent

coding & agent channel →