d1.2-ultra is a high-performance AI decision model created by Liquid AI to evaluate unstructured information and generate accurate, structured predictions. Unlike conventional generative language models, it focuses on answering specific questions and returning probabilities, classifications, extracted values, or numerical estimates. The model supports multimodal inputs, including written text, images, and audio recordings, allowing applications to analyze information from different sources. Its document extraction capabilities identify specific text passages, numerical values, dates, and personally identifiable information while preserving references to their original locations. Visual analysis tools can detect objects, return bounding boxes, count occurrences, and estimate quantities within a defined range. Audio understanding enables applications to recognize spoken languages, identify environmental sounds, and interpret questions submitted through voice recordings. d1.2-ultra also supports yes-or-no evaluations, single-choice and multiple-choice classification, and numerical scoring with calibrated probability distributions. According to Liquid AI, the model achieved a leading score of 75 on the Artificial Analysis Decision Index v2.1 and demonstrated a calibration error of 0.8 percentage points across its evaluated benchmarks. Through the Liquid API, developers can integrate d1.2-ultra into document automation, fraud detection, customer service routing, compliance monitoring, visual inspection, and other decision-oriented applications.