I'm afraid of spiders. So I made AI look at 2k of them (labqoat.com)

In a unique exploration of AI capabilities, a researcher tested nine models to identify 2,000 spider photos, examining how accurately they could classify these creatures into species. Using images sourced from iNaturalist observations, the study aimed to benchmark AI performance in distinguishing within a taxonomy that requires nuanced recognition. Each model received the same selection of photos alongside a multiple-choice list of potential species names to determine their accuracy, with results indicating varying success rates. Significantly, the findings demonstrated that while no model achieved a perfect score, “Gemini 3.8 Flash” emerged as the leader, correctly identifying nearly 50% of the species. Other models, like "GPT-6 Astra" and "Claude Fable 5.1," followed but at a greater cost, emphasizing a notable trend where efficiency and accuracy can be balanced against financial investment in AI tools. The results reveal not only the challenge of precise biological classification but also the potential for AI to assist in real-world applications in species identification—an area of growing importance in field biology and biodiversity conservation.