The new breakthrough in AI species identification: 95,000 species recognized

The BioCLIP 2 model developed by The Ohio State University, trained on 200 million biological images, can identify 950,000 species and understand ecological relationships and individual differences without supervised signals. The model achieves a zero-shot accuracy of 55.6% in species identification, outperforming the second-place SigLIP model by 16.1 percentage points. In non-species visual tasks—such as habitat recognition and plant disease identification—BioCLIP 2 also demonstrates strong performance. Feature space analysis reveals that emergent properties, including ecological alignment across species and separation of intra-species variations, improve with increasing training scale, further confirming the model’s ability to learn biological knowledge through large-scale training.