[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-ricursive-ripple-65b-dense-open-weight":3,"topics-all":38,"news-related-4193fdcb-58a9-484e-9bd7-e4071c6588bd":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},"4193fdcb-58a9-484e-9bd7-e4071c6588bd","Ripple-65B:65B稠密模型80.52分压过975B","美国实验室 Ricursive 发布 65B 稠密模型 Ripple-65B：12 万亿 token 预训练、100 万上下文，官方自报终端基准 80.52 分，称刷新美国开放权重模型在智能体编码上的成绩。总研发成本 3200 万美元，约为此前榜首的十分之一，但全部跑分均由厂商自己测得，第三方复现尚无。","美国时间 11 月 4 日，Ricursive Intelligence 放出一份和主流反着来的权重。它不是又一个把参数堆到上千亿的 MoE，而是 65B 稠密 Transformer，全部 65B 参数每个 token 都激活。官方模型卡自报 TerminalBench 2.1 拿 80.52 分，同一张表里被它压过的是 Reflection Beam 501B（80.1）、Meta Muse Spark Open 816B（76.4）、Thinking Machines Inkling 975B（63.8）和 NVIDIA Nemotron 3 Ultra 550B（56.4）——总参数量大约是它的 8 到 15 倍。要说明的是，这组对比数字全部出自 Ricursive 自己的表格，[模型卡](https:\u002F\u002Fhuggingface.co\u002FricursiveAI\u002FRipple-65B-Preview)写明缺失的评测由自家 in-house 流程补齐。\n\n## 65B 打赢 975B，靠的不是参数量\n\n真正被挑战的不是参数量，是路线。当下的开放权重模型几乎清一色走 MoE：把总参数做大，用稀疏激活把每次推理的算力压下来，代价是模型复杂度和训练基础设施都跟着上去。Ripple 选择了相反的路线，官方给出的理由很直接：稠密模型换来的是低延迟的 token 生成，以及训练与服务两侧都更简单的工程复杂度。这条路的赌注很直白——让全部参数都参与进来，换一份更简单、训练和服务都不那么贵的模型。\n\n## 模型是为跑它的那台机器设计的\n\n结构不是随手选的。Ripple 走 thin-and-deep：d_model 只有 5120，层数堆到 80，从 Olmo 3 的 llama-like 布局换成 Gemma 3N 风格以稳住大规模训练；注意力头刻意超配，同时提高滑窗层占比、缩小滑窗窗口来抵掉成本，再叠上 gated attention、global NoPE、value residuals 和 FFN activation clipping。结果是全程 40% MFU，只用简单分片，没有自定义 kernel，也没做量化——团队规模从 1 人长到 5 人。推理侧，2 张 B200、batch=8 在量化前能跑 250 tok\u002Fs；65B 能放进 NVIDIA DGX Spark，官方另发了 NVFP4 版本和一份上手 colab。训练配方是从 Olmo 3 的模型阶梯改出来的，一路迭代到 v1.6：中途也试过 2-stream mHC，因为在 4B 规模上没看到收益被直接否掉。完整设计与取舍写在[发布博客](https:\u002F\u002Fwww.ricursive.com\u002Fripple)。\n\n## 3200 万美元和 140 万次 rollout\n\n账本比跑分更值得看。整个研发——预训练阶梯实验、数据配比、评测、合成数据、后训练配方，直到最终训练——官方称花了 3200 万美元，约为此前开放权重智能体编码榜首的十分之一。这是公司自己的估算，参照系是 Reflection 被报道每月 1.5 亿美元的算力承诺。后训练只用了 140 万次 RL rollout 就打到 80.5，官方称约是 Reflection Beam 的 1\u002F70，而后者被报道用过超过 1 亿次。后训练各阶段的 token 分配也摊在模型卡上：中期训练 200B、长上下文 6.5B、SFT 183B、长上下文 SFT 13.5B、RL 700 步，长上下文阶段直接按 2M 序列长度训。预训练侧从 Olmo 3 的阶梯做起，4B 规模上代码算力效率提升 16 倍，完整 hero run 上最高到 55 倍。\n\n## 榜单得打折看\n\n换成官方 Terminus 2 harness、128K 上下文重跑，竞品普遍掉分：Reflection Beam 从 80.1 掉到 70.8，Inkling 63.8 掉到 56.2，Nemotron 3 Ultra 56.4 掉到 55.1，Muse Spark Open 反而升到 78.7，仍低于 Ripple 的 80.5。弱项也有：SciCode 39.4 在五个模型里垫底，SWE-bench Verified 76.5 低于 Beam 的 81.6 和 Inkling 的 77.6，MCP Atlas 71.67 同样排在后面。中国开放权重模型不在这张表里；据 IAMagazine 核查，发布时 Ripple 也尚未进入 Artificial Analysis 的 Intelligence Index v5，第三方独立复现目前没有。\n\n## 所以呢\n\n真正的看点不是又一款更大的模型，而是「小而稠密 + 本地跑得动 + 成本可核对」这条路线第一次被完整摊开：架构、训练配方、硬件适配、每一步花多少钱都写在明面上，可以被人逐项查。TerminalBench 这类智能体基准对评测脚手架格外敏感，同一批模型换个 harness 分数就能整体位移，这 80.52 分该打折看。但如果路线成立，受影响的不只是终端编码这一张榜单，而是「默认必须用 MoE 才能追前沿」这个前提本身。","https:\u002F\u002Fwww.ricursive.com\u002Fripple","da034c50-2c33-42f1-8113-80e51f045fc3",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"120fa59a-ff6f-4537-9bf5-f818df636a0e","benchmark",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":19,"name":20,"slug":20,"description":14,"color":14},"7e89b5cc-57db-4f37-bc6d-28919a73931c","model-release",{"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},"a35763c2-30be-446f-b248-f1487fff8943","en","Ripple-65B: 65B dense model tops 975B MoEs on agentic coding","Ricursive Intelligence's Ripple-65B Preview is a 65B dense Transformer, 12T tokens, 1M context, scoring 80.52 on TerminalBench 2.1 for 32M USD in R&D.","On November 4, Ricursive Intelligence released a set of weights that runs against the grain of the current industry. It is not another MoE scaled up to hundreds of billions of parameters, but a 65B dense Transformer with all 65B parameters active on every token. The official model card reports 80.52 on TerminalBench 2.1. The comparison table on that same card lists the models it is said to pass: Reflection Beam 501B at 80.1, Meta Muse Spark Open 816B at 76.4, Thinking Machines Inkling 975B at 63.8, and NVIDIA Nemotron 3 Ultra 550B at 56.4 — models carrying roughly eight to fifteen times its total parameter count. It is worth being precise about where these numbers come from: the entire comparison is the Ricursive table, and the [model card](https:\u002F\u002Fhuggingface.co\u002FricursiveAI\u002FRipple-65B-Preview) states that missing evaluations were filled in using an in-house eval flow.\n\n## 65B beating 975B, and not on parameter count\n\nWhat is really being challenged is not parameter count but route. Nearly every open-weight model today goes MoE: grow the total parameter count, use sparse activation to keep per-inference compute down, and accept that both model complexity and training infrastructure get harder. Ripple took the opposite route, and the stated reason is straightforward. A dense model buys low-latency token generation and much simpler engineering on both the training and serving side. The bet is plain: let every parameter participate, and get a model that is simpler to train and cheaper to serve.\n\n## The model was designed for the machine that runs it\n\nThe architecture was not picked at random. Ripple takes a thin-and-deep shape: a d_model of only 5120 spread across 80 layers, moving from Olmo 3 llama-like layout to a Gemma 3N-like one to keep training stable at scale. Attention heads are deliberately overprovisioned, while the sliding-window-to-global layer ratio goes up and the window size goes down to compensate for the cost. On top of that sit gated attention, global NoPE, value residuals, and FFN activation clipping. The result is a solid 40 percent MFU throughout pretraining and posttraining, achieved with simple sharding, no custom kernels, and no quantization — by a team that grew from one person to five. On the serving side, two B200s at batch size 8 reach 250 tokens per second before quantization, and the 65B model fits on an NVIDIA DGX Spark. A separate NVFP4 checkpoint and a getting-started colab are also published. The training recipe started from the Olmo 3 model ladder and was iterated up to v1.6, with a 2-stream mHC variant tried and dropped because it showed no gain at 4B scale. The full design rationale and the accepted and rejected trade-offs are in the [launch blog](https:\u002F\u002Fwww.ricursive.com\u002Fripple).\n\n## Thirty-two million dollars and 1.4 million rollouts\n\nThe accounting is more revealing than the scores. For the entire effort — pretraining ladder experiments, data mixing, evaluations, synthetic data, posttraining recipe development, and the final training runs — the company says it spent 32 million dollars, roughly one tenth of the previous open-weight leader on agentic coding. That comparison is the company own estimate, benchmarked against a reportedly 150 million dollars a month compute commitment by Reflection. Posttraining reached 80.5 using only 1.4 million reinforcement learning rollouts, which the company puts at around one seventieth of what Reflection Beam used, the latter reportedly running past 100 million. The token budget per posttraining stage is also spelled out on the model card: 200B for midtraining, 6.5B for long context, 183B for SFT, 13.5B for long-context SFT, and 700 RL steps, with the long-context stages trained at a 2M sequence length. On the pretraining side, the recipe started from the Olmo 3 ladder and improved code compute efficiency by 16x at 4B model scale, reaching as much as 55x for the full hero run.\n\n## Discount the leaderboard\n\nRe-run under the official Terminus 2 harness at 128K context and the rivals mostly fall: Reflection Beam drops from 80.1 to 70.8, Inkling from 63.8 to 56.2, and Nemotron 3 Ultra from 56.4 to 55.1, while Muse Spark Open actually rises to 78.7 — still below Ripple at 80.5. There are weak spots too. SciCode at 39.4 is the lowest of the five models, SWE-bench Verified at 76.5 trails both Beam at 81.6 and Inkling at 77.6, and MCP Atlas at 71.67 also sits behind. Chinese open-weight models are not part of the comparison at all, and IAMagazine checked that Ripple had not yet appeared in the Artificial Analysis Intelligence Index v5 at release time. No third-party independent replication exists yet.\n\n## So what\n\nThe interesting part is not yet another larger model. It is that the path of small-and-dense, runs-locally, and cost-auditable has been laid out in full for the first time: architecture, training recipe, hardware adaptation, and how much each step cost, all written down where anyone can check them line by line. TerminalBench-style agentic benchmarks are unusually sensitive to the evaluation harness, and the same set of models can shift as a whole when the harness changes, so that 80.52 deserves a discount. But if the approach holds, what is affected goes beyond a single terminal coding leaderboard to the premise that you need Mixture-of-Experts to reach the frontier at all.","ricursive-ripple-65b-dense-open-weight","2026-11-05T05:30:00Z","2026-11-05T05:20:23.334239Z","2026-11-05T05:20:23.334249Z",true,"agent",175,[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},"d529114c-bfd7-4efb-9af4-444d3065b397","Step 5 Preview 开源,2.7 美元\u002F百万输出","stepfun-step-5-preview-600b-moe","2026-10-17T21:12:04+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"5a5efb41-7037-46cb-b93e-6842f9ff2d62","UltraBERT:91.0 GLUE+47.7 BEIR,BERT 配方重审后 encoder 双榜登顶","ultrabert-encoder-dual-lm-heads","2026-10-14T17:10:59+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"4c4a2a9e-f69b-4985-bd42-97ab2ef4e2ac","Spark-X2.5-4B 开源:4B 跑 1M 上下文,22 项基准打 9B 级 Qwen3.5","spark-x2-5-4b-apache-open-source","2026-09-16T01:30:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"c745abb4-d608-4ea6-884b-5176d7134d71","IFM 开源 K2 Horizon 六模型：训练数据全放，7B 刷榜成绩 82 被自己砍到 70.6","ifm-k2-horizon-open-fleet-audit","2026-09-05T23:07:55+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"741bd34c-7134-4e8e-ab45-4f53dc576a6b","腾讯 Hy4 登顶 9 月开源榜:79.87 分超 Qwen3.8 Max,Anthropic 包揽总榜前三","tencent-hy4-tops-open-source-benchlm-september","2026-09-01T17:10:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"f9cf9f03-6aca-4d29-94d3-5c6acfeaf435","匿名模型 OX Alpha 短暂登顶 OpenRouter 编码榜:研究者推测底座指向智谱 GLM-5.x","ox-alpha-stealth-openrouter-glm-5-zhipu","2026-08-24T03:00:00+00:00"]