d1-3B is a 3.12B-parameter decision model from Liquid AI, post-trained from the LFM2.5-VL-3B vision-language model. It is not a chat model and does not write text: you give it a state (text, JSON, images or a mix) and a set of named questions, and it returns typed, calibrated answers in a single forward pass with zero output tokens. Question types are yes/no, choice among named options, and score over 2 to 10 ordered levels, with confidence and probabilities returned for choice and score. Suggested uses include routing and triage, moderation, intent classification, extraction checks, reranking, LLM-as-a-judge scoring, agent guardrails and visual inspection. The model pairs a 400M SigLIP2 NaFlex vision encoder with a 32,768-token context and loads through transformers (>=5.14) with trust_remote_code. On Decision Index 0.2.1, scored with the official scorer rather than submitted to the leaderboard, it reaches 48.57, compared with 50.02 for the 12B Winnow-12B and 47.11 for Decider 35B-A3B. Across 11 public image benchmarks read as decisions it averages 74.1 against 73.9 for the base model, with gains on VL-RewardBench (65.0 vs 50.9) and MMVP (77.0 vs 73.7) and losses on CV-Bench (82.1 vs 87.6) and VisualWebBench (71.4 vs 78.3); with the images removed the same questions score 45.1. Reported latency for one question is 8 ms on an RTX 4090 and 30 ms on an Apple M5 Pro. Weights ship under the LFM Open License v1.0, which grants use and redistribution but does not license commercial use by a legal entity with annual revenue of 10 million US dollars or more.