According to CoinMeta, the multimodal decision-making model d1-ultra released by Liquid AI scored 65 points in the decision, index, v1.0 of artificial, surpassing the previous generation's 57 points by 1.14 points to rank first. This model is not responsible for chatting or generating articles; instead, it reads text and images to directly determine which option is more likely to be correct and provides a probability for each. For example, it can inspect circuit board defects, verify product images against descriptions, or review claims photos uploaded by customers. However, in a 500-question knowledge test in the artificial and analysis, d1-ultra only answered correctly about 32%, which is lower than the 41% of the previous generation d1. Its advantage lies in answering less randomly, with the proportion of choosing "I don't know" increasing from 2% to 85%. d1-ultra has already passed the liquid, api, and vercel evaluations, and the model's weights have not yet been made public. It charges $0.042 per million inputs, which is 2.1 times that of the previous generation d1, and there are no fees for output.