[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-liquid-ai-d1-open-edge-decision-models":3,"topics-all":38,"news-related-df4159eb-60c1-4267-835b-0932d2a1ac72":57},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"df4159eb-60c1-4267-835b-0932d2a1ac72","d1-3B 开源：零token决策，8毫秒应答","Liquid AI 开源 d1-3B 与 d1-omni-600M 决策模型：一次前向直接返回带置信度答案，零输出 token，RTX 4090 上 8 毫秒。d1-3B 在 Decision Index 拿 48.57 分领先全部 10B 以下模型，把决策模型拉进摄像头逐帧时代。","决策模型赛道最近两周挤满大厂：微软、亚马逊、Cloudflare、Perplexity 先后开源自家模型，打法一致：把 LLM 路由压成一次廉价分类。Liquid AI 这次的东西不太一样：多模态、逐帧、零 token 输出。\n\n## 不生成 token 的模型\n\nd1-3B 与 d1-omni-600M 属于 Liquid AI 的 d1 决策模型家族，参数量 3.1B 与 600M，分别从 LFM2.5-VL-3B 和 LFM2.5-Encoder-350M 后训练而来。它与生成式 LLM 的根本区别：不吐 token。给它一个状态（文本、JSON、图片或混合）和一组问题，一次前向传播直接返回带置信度的类型化答案——是非、多选一、评分档位，接口用量字段里 output_tokens 恒为 0。\n\n延迟因此进入生成式模型够不着的区间：RTX 4090 单问题 8 毫秒，Jetson AGX Thor 16 毫秒，最小的 Orin Nano 也只要 50 毫秒；64 状态并发打包，AGX Thor 每秒吞 262 次。官方还与 NVIDIA 在 Isaac Sim 做了硬件在环演示，模型跑在 Jetson 板上直接导航。\n\n## 跑分：3B 排到全榜第二\n\nDecision Index v0.2.1 公开榜上，d1-3B 拿 48.57 分，仅次于 12B 的 Winnow-12B（50.02），也压过 36B 的 Decider 35B-A3B（47.11）——体积约 12 倍于它。分项里 Tools 74.5、Arts 36.3 均为榜单最高。七项公共文本基准平均 82.9，高于 Decider 4B 的 81.1；d1-omni-600M 平均 78.4，以约四分之一参数打赢 Decider 2B，毒性检测与改写识别两项全表最高。\n\n## 训练复盘：朴素技巧赢了\n\n官方复盘最有意思的一句：长输入训练、打乱答案选项、修掉数据捷径，「比更高级的技巧作用更大」。d1-3B 的底座是把 LFM2.5-2.6B 与 LFM2.5-VL-3B 文本骨干的权重直接平均，再用不同种子与数据配比微调多个 checkpoint 合并。omni 版分阶段加装：音频走 FastConformer 加 adapter，视觉取 LFM2.5-VL-450M 编码器配 LoRA。\n\n## 多模态是差异点，短板也没藏着\n\n11 项视觉基准平均 74.1，对底座 73.9 基本持平：MMVP 从 73.7 升到 77.0，VL-RewardBench 从 50.9 涨到 65.0；CV-Bench 从 87.6 回落到 82.1——官方把升降都摆进了同一张表。删掉图片同题只剩 45.1 分，答案确实读自图像。omni-600M 是 600M 级少见的文+图+音三模态组合，但 15.95 的分数被官方标注为实验性 checkpoint；音频决策基准「仍是开放问题」。\n\n配套 10 个 demo 跑在 Hugging Face Space，摄像头逐帧调用 d1-3B，每帧读一次答案。llama.cpp 首日支持，权重开放下载。\n\n## 所以呢\n\n两周内五家同押决策模型，「把路由和分类从 LLM 拆出来」已是共识。但前四家的叙事都围绕文本路由性价比，d1-3B 的真正增量是把「每一帧都可以问一次模型」变成工程现实——50 毫秒的 Orin Nano 意味着决策模型可以嵌进摄像头流水线逐帧工作，这是生成式 LLM 结构上做不到的事。3B 换 12B 的榜一差距，在边缘场景怎么算都不亏。\n\n参考：[官方博客](https:\u002F\u002Fwww.liquid.ai\u002Fblog\u002Fd1-open) · [模型卡](https:\u002F\u002Fhuggingface.co\u002FLiquidAI\u002Fd1-3B)","https:\u002F\u002Fwww.liquid.ai\u002Fblog\u002Fd1-open","e20eb2f5-92d2-475d-bfe6-6785c043169e",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"fca9258a-9430-455a-b95d-b9fae5e373a8","ai-inference",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"7e89b5cc-57db-4f37-bc6d-28919a73931c","model-release",{"id":19,"name":20,"slug":20,"description":14,"color":14},"499f4b56-819d-49a3-9609-33e775143b86","multimodal",{"id":22,"name":23,"slug":23,"description":14,"color":14},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"97476cea-699c-476b-a7c5-83d568d29d16","en","Liquid AI Open-Sources d1-3B: Zero-Token Decisions in 8ms","Liquid AI open-sources d1-3B and d1-omni-600M, decision models that answer in one forward pass with zero output tokens — 8ms on an RTX 4090.","The decision-model lane has been crowded lately: within two weeks Microsoft, Amazon, Cloudflare and Perplexity all open-sourced their own models, with an identical playbook — compress LLM tool-call routing into one cheap classification. What Liquid AI just shipped is different: multimodal, per-frame, zero-token output.\n\n## A model that generates no tokens\n\nd1-3B and d1-omni-600M belong to Liquid AI's d1 decision-model family — 3.1B and 600M parameters, post-trained from two different backbones, LFM2.5-VL-3B and LFM2.5-Encoder-350M. The fundamental difference from a generative LLM: it emits no tokens. You hand it a state (text, JSON, an image, or a mix) plus a set of questions, and a single forward pass returns calibrated, typed answers — yes\u002Fno, a choice among named options, a rating level. The official API's usage field reports output_tokens as exactly 0.\n\nThat design pushes latency into territory generative models cannot reach: 8 ms per single question on an RTX 4090, 9 ms on an AMD MI325X, 16 ms on a Jetson AGX Thor, and 50 ms even on the smallest Jetson Orin Nano. Packing three questions into one forward pass moves the AGX Thor from 16 ms to just 20 ms; with 64 states packed together, the AGX Thor processes 262 per second. In collaboration with NVIDIA, there is also a hardware-in-the-loop demo in Isaac Sim — the model runs on a Jetson board and makes navigation decisions directly.\n\n## Benchmarks: a 3B model ranked second overall\n\nOn the public Decision Index v0.2.1 leaderboard, d1-3B scores 48.57 — second only to the 12B Winnow-12B (50.02), ahead of every other sub-10B model, and above the 36B Decider 35B-A3B (47.11), a model roughly 12x its size. In the sub-scores, Tools at 74.5 and Arts at 36.3 are the highest on the board.\n\nAcross seven public text benchmarks (SQuAD 2.0, Civil Comments, MASSIVE intent, PubMedQA, BoolQ, XNLI, PAWS-X), d1-3B averages 82.9, above Decider 4B (81.1) and Decider 2B (77.1). The smaller d1-omni-600M averages 78.4, beating Decider 2B with about a quarter of the parameters; its toxicity-detection score (Civil Comments 95.8) and paraphrase-identification score (PAWS-X 79.5) are the highest in the table.\n\n## Training post-mortem: plain tricks won\n\nThe most interesting line in the official write-up: training on long inputs, shuffling answer options, and fixing shortcuts in the data \"made a bigger difference than more advanced techniques.\" The d1-3B base itself comes from directly averaging the weights of LFM2.5-2.6B and the text backbone of LFM2.5-VL-3B, then fine-tuning multiple checkpoints with different random seeds and data mixtures before merging them again. The 600M omni variant added modalities in stages: audio via a FastConformer encoder with an adapter, vision borrowed from the LFM2.5-VL-450M encoder with LoRA updates — active only when the input includes images, with the vision encoder frozen throughout.\n\n## Multimodal is the differentiator, and the gaps are stated plainly\n\nAcross 11 public vision benchmarks, d1-3B averages 74.1 versus the backbone's 73.9 — essentially flat: MMVP rises from 73.7 to 77.0 and VL-RewardBench jumps from 50.9 to 65.0, but CV-Bench falls from 87.6 to 82.1 and VisualWebBench from 78.3 to 71.4. The company published both directions in the same table. With the images removed, the same questions drop to 45.1, showing the answers genuinely come from the images. The omni-600M is a rare text+image+audio triple-modality combination at the 600M scale, but its 15.95 Decision Index score is explicitly labeled an experimental early checkpoint, and dedicated audio decision benchmarks \"are currently an open problem\" — a gap left to the community.\n\nAll 10 companion demos run on a Hugging Face Space, calling d1-3B frame by frame from a live camera: gesture-controlled games, real-time content moderation, one answer read per frame. On the inference side, llama.cpp is supported from day one across Apple, AMD, Qualcomm and NVIDIA (including NVFP4), and the weights are open for download.\n\n## So what\n\nFive players — Microsoft, Amazon, Cloudflare, Perplexity and now Liquid AI — have bet on decision models within two weeks, which says that \"splitting routing and classification out of the LLM\" is now industry consensus. But the first four all frame it around text-routing cost-performance and leaderboard scores; the real increment from d1-3B is making \"you can ask the model once per frame\" an engineering reality — a 50 ms Orin Nano means a decision model can sit inside a camera pipeline working frame by frame, something a generative LLM structurally cannot do. Trading 3B for a 12B-wide gap to first place looks fine anywhere edge deployment matters. The next thing to watch is whether audio decision benchmarks emerge — that is where the omni variant would truly belong.\n\nSources: [Liquid AI blog](https:\u002F\u002Fwww.liquid.ai\u002Fblog\u002Fd1-open) · [d1-3B model card](https:\u002F\u002Fhuggingface.co\u002FLiquidAI\u002Fd1-3B)","liquid-ai-d1-open-edge-decision-models","2026-10-11T19:14:50Z","2026-10-11T19:14:53.732833Z","2026-10-11T19:14:53.732848Z",true,"agent",7,[39,48],{"slug":40,"tag_slug":40,"title_zh":41,"title_en":42,"intro_zh":43,"intro_en":44,"id":45,"is_active":35,"created_at":46,"modified_at":47},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":49,"tag_slug":49,"title_zh":50,"title_en":51,"intro_zh":52,"intro_en":53,"id":54,"is_active":35,"created_at":55,"modified_at":56},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":58},[59,64,69,74,79,84],{"id":60,"title":61,"news_slug":62,"published_at":63},"bcdc10bc-2f08-4c39-8ffa-e7e34041c112","京东开源 JoyAI-Video-Edit:用 16B 多模态扩散 Transformer 把视频编辑推进「边播边改」实时流时代","jd-joyai-video-edit-real-time-streaming","2026-08-05T03:00:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"c53a7c3b-7a0c-4cc6-822e-aa0d66e01abe","EmbeddingGemma 2 开源:740M 多模态嵌入跑进手机","embeddinggemma-2-multimodal-embeddings","2026-10-11T21:10:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"36c71e4e-e398-4583-8710-1732aefff06a","OneStreamer:4B 流式模型先记再答,八榜最佳","onestreamer-4b-streaming-video-memory","2026-10-02T15:07:48+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"bbe8d55a-1069-42fa-a342-d944d53fdb4b","操作电脑的 27B 开源权重模型 Holo4:最强版禁商用","holo4-open-weight-computer-use","2026-10-02T13:12:01+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"dc61debd-1d77-4d5a-9d59-5b23c3da07de","蚂蚁开源Realtime-Venus：9B全双工模型边说边干活，三项续聊指标超GPT-4o","ant-realtime-venus-full-duplex-delegation","2026-09-30T23:10:53+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"0d357a0b-42da-40af-8038-e8c035cb9810","Apple 开源 LensVLM-9B:先扫压缩图,再读原页","apple-lensvlm-9b-weights-huggingface","2026-09-26T15:20:00+00:00"]