InterLV-Search benchmark shows current multimodal agents achieve below 50% accuracy on interleaved language-vision search tasks involving repeated evidence use and multi-branch comparisons.
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
-
InterLV-Search: Benchmarking Interleaved Multimodal Agentic Search
InterLV-Search benchmark shows current multimodal agents achieve below 50% accuracy on interleaved language-vision search tasks involving repeated evidence use and multi-branch comparisons.