Reproduce “How Finely Can Vision-Language Models See Biology? Bio-VQA: A Benchmark for Taxonomic, Morphological, Spatial and Individual Reasoning” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

S M Rayeed, Qian Ma, Charles V. Stewart, Yao Ma

The authors introduce Bio-VQA, a benchmark of 1,861,608 questions on 216,715 images from 11 datasets, covering 24 biological vision tasks. They evaluate 18 open-weight models in zero-shot, single-pass settings and compare biological performance with image-blind floors. Qwen3-VL-32B counts animals exactly in the 2–10 range on 41.7% of single-image items, versus 94.9% for everyday objects, a benchmark-specific gap.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.