[KongchangAI]
BenchmarkExtractBench

ExtractBench

A benchmark for evaluating structured information extraction capabilities.

Timeline (last 90 days)

Oct 13

NuExtract3 scored 0.651 in NuMind's internal test of about 600 documents, versus 0.417 for the base model Qwen3.5-4B; on ExtractBench it averaged 82.35, but neither states coverage of financial report documents.

Unverified50%

All Facts (1)

Source Articles