BloomBench reveals that state-of-the-art VLMs perform well on semantic understanding but struggle with factual recall and creative synthesis, while also showing large English-Arabic performance gaps.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Almieyar-Oryx-BloomBench: A Bilingual Multimodal Benchmark for Cognitively Informed Evaluation of Vision-Language Models
BloomBench reveals that state-of-the-art VLMs perform well on semantic understanding but struggle with factual recall and creative synthesis, while also showing large English-Arabic performance gaps.