From Pixels to Pairs: A Comprehensive Benchmark of LLM-Based Key-Value Extraction in Noisy Document Settings

What changed

New benchmark shows LLM key-value extraction performance degrades substantially under OCR noise, narrowing model gaps.

Why you should care

LLM extraction performance degrades substantially under OCR noise, narrowing model gaps.

Your move

Test. OCR noise dominates performance; test extraction pipelines with realistic OCR inputs.

Evidence

ConfirmedarXiv — cs.CL ↗

What to watch next

Release of updated OCR models or LLMs specifically fine-tuned for noisy document extraction tasks.

Event

research

Event date

Sep 17, 2026

Relevant to

General AI readers