The authors evaluated OpenVLHarness on 22 benchmarks covering grounded perception, visual search and deep research, visual question answering, hallucination and spatial reasoning.
Visual agents improve when tool evidence persists across reasoning steps
OpenVLHarness gives multimodal models consistent access to external tools while retaining images, data and web sources for later computation.
Big Tech
Yu Zhou · Cheng-Fu Yang · Rui Sun · Di Wu · Sibo Peng · Mingyang Zhang · +5 more
University of California, Los Angeles · Apple
Research Digest··2 min read
Zhou et al.
§