When NHSPID first launched in July 2025, it knew exactly 36 species, all of them jumping spiders. The idea was simple: could a model trained on community photos learn to tell New Hampshire's jumping spiders apart? It could, at around 81% accuracy.

v2 is a much bigger project.

Far beyond the jumping spiders

v2 recognizes 161 species, and it's no longer just jumping spiders. We added them family by family: orb-weavers, wolf spiders, crab spiders, cobweb spiders, cellar spiders, fishing spiders, and others. If you've spotted a spider in New Hampshire, there's a good chance v2 now knows it.

A new set of eyes

v2 runs on a new backbone, a DINOv2 Vision Transformer (ViT-L/14), and looks at each photo at 518 pixels with around 300 million parameters. Even though the number of species more than quadrupled, accuracy on held-out test photos went up, from 81.4% to 91.7%.

Built on community observations

It all comes from the community. v2 learned from 123,099 photos across 60,859 research-grade observations on iNaturalist, each one uploaded by someone who stopped and took a closer look. We use New Hampshire to decide which species belong on the list, but the training photos come from across each species' range, wherever people happened to be observing. A species made the list once it had at least 30 observations and 75 photos to learn from, and we only used openly licensed images.

Give it a try

Head over to the Identify page and upload a photo, or browse the full list of 161 species the model now knows. And if you're out this summer, keep observing and uploading your photos to iNaturalist. Researchers everywhere rely on that data, and every photo also helps the next version of NHSPID see a little better.