Liquid AI’s new d1-ultra model has topped the decision-making leaderboard, surpassing Jev, while its accuracy on knowledge questions dropped to 32%.

Insight Beating AI Flash News: Liquid AI has released the multimodal decision model d1-ultra, which scored 65 points in Artificial Analysis’s Decision Index v1.0, surpassing Jev 1.14’s 57 points to claim the top spot. Unlike models built for chatting or article generation, this decision model reads text and images to directly judge which option is more likely correct and provides corresponding probabilities. Its applications include inspecting circuit board defects, verifying that product images match their descriptions, or reviewing customer-uploaded claim photos. However, the new model’s top ranking comes with a notable tradeoff: in Artificial Analysis’s 500-question knowledge test, d1-ultra answered only ~32% of questions correctly, lower than the previous d1 generation’s 41%. Its strength lies in reducing incorrect responses. Among questions it did not answer correctly, the share of choosing “I don’t know” rose from 2% to 85%. When facing uncertain queries, it is more willing to admit ignorance rather than risk providing wrong answers — a key factor behind its significant ranking improvement. Artificial Analysis assigns 12.5% weight each to knowledge accuracy and the ability to avoid incorrect responses, with the remaining score derived from complex reasoning, tool call judgment, and safety classification. Liquid AI’s own 112-test evaluation shows d1-ultra outperforms Jev 1.14 across all eight task categories. d1-ultra is available via Liquid API and Vercel; its model weights have not been open-sourced. It charges $0.042 per million input tokens, 2.1 times the price of the previous d1 model, with no output fees. For a single image used to answer multiple questions, input costs are recalculated for each question.