{"slug":"liquid-d1","name":"Liquid AI d1-3B","nameZh":"Liquid d1（端侧决策模型）","category":"models","categoryName":"Open-weight models","audience":["ai-era"],"primary":"ai-era","openClosed":"open","pricing":"open-source","license":"LFM Open License v1.0","maintained":"yes","officialUrl":"https://huggingface.co/LiquidAI/d1-3B","repo":null,"latestVersion":null,"latestRelease":null,"stars":null,"starsAsOf":null,"description":"Liquid AI's 3B multimodal decision model built on LFM2.5-VL-3B: give it a state (text/JSON/images) plus yes-no, choice or score questions and it returns calibrated probabilities in one pass with zero output tokens; ~8 ms per decision on an RTX 4090, ~30 ms on Apple M5 Pro. For routing, moderation, intent, agent guardrails — not a chat model","descriptionZh":"Liquid AI 基于 LFM2.5-VL-3B 的 3B 多模态“决策模型”：给状态（文本/JSON/图片）和一组 是否/选择/打分 问题，一次前向直接给出校准概率、0 输出 token；RTX 4090 约 8 ms、Apple M5 Pro 约 30 ms 一次决策，适合路由、审核、意图分类、agent 护栏。不能聊天写文本","callMethods":[],"platforms":[],"addedAt":"2026-10-14","updatedAt":null,"topics":[],"language":null,"lastPush":null,"formerly":null,"packages":[],"recentVersions":[],"alternatives":["deepseek-v3","qwen3","gpt-oss","qwen3-coder","kimi-k2","llama-models","ultralytics","gemma"],"stacks":[],"scenarios":[],"problems":[],"reasonZh":null,"sources":["editorial"],"record":"https://aimap.page/api/v1/software/liquid-d1.json","pages":{"en":"https://aimap.page/en/software/liquid-d1/","zh-hans":"https://aimap.page/zh-hans/software/liquid-d1/"}}