ConfBench
Explore score history ↗ConfBench is a calibration-specific benchmark for key information extraction from documents. It applies 20 degradation pipelines to create 1,346 variants and over 70K entity-level evaluations, spanning the accuracy spectrum for evaluating confidence estimates of VLMs.
- Released
- 2026-08-03
- Evaluates
- General AI, Multimodal
- Openness
- unknown
- Importer
- Claire Radar
- Review status
- unreviewed
ConfBench paper
No reported scores are on record for this benchmark yet.