[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-beam-1-0-weights-open-agent-stack":3,"topics-all":38,"news-related-bbfcbee4-8d92-4078-bb1e-eaf4d1c1e146":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},"bbfcbee4-8d92-4078-bb1e-eaf4d1c1e146","Beam 权重落地:501B MoE 三精度齐发","Reflection AI 首个开源权重模型 Beam 1.0 落地:501B\u002F23B 稀疏 MoE,Apache 2.0,BF16\u002FFP8\u002FNVFP4 三精度齐发,含投机解码 drafter。官方称 token 用量约为同级三分之一,40 余家伙伴提供 day-0 支持。","10 月 16 日,Reflection AI 把 Beam 1.0 的模型权重放上了 Hugging Face:501B 总参数、23B 激活的稀疏 MoE,Apache 2.0 许可,BF16、FP8、NVFP4 三种精度连同投机解码 drafter 一起发布。距离官宣仅 11 天,这家公司的第一个开源权重模型从「发布预告」进入「可下载、可自托管」的阶段。\n\n## 架构:1M 上下文的稀疏 MoE\n\n按 model card,Beam 总参数 501.3B、激活 23.4B,52 层里 3 层 dense、49 层 MoE,共 256 个专家;注意力用 Gated GQA 加滑窗机制,上下文 1M token,词表 152k,聊天模板是改造过的 Harmony 格式,知识截止 2026 年 6 月 30 日。它只做文本,定位编码、推理和长程 Agent 任务。\n\n官方把 token 效率列为第一能力:model card 称 Beam 用约少 3 倍的推理 token 就能达到与领先开源模型相当的解题质量。\n\n## 基准:三精度差距很小\n\n评测统一在 temp=0.7、top_p=0.99 下跑。代表数字(BF16):SWE Bench Verified 81.6%、SWEBench Multilingual 78.3%、Terminal Bench v2.1 80.1%、AIME 2026 97.7%、GPQA Diamond 90.6%、LiveCodeBench 88.3%。低精度并未明显掉点:SWE Bench Pro v2-Hard 一项,NVFP4 版 78.6% 反而高于 BF16 的 76.5%;全表三精度差距大多在两个百分点上下。\n\n短板也直接写在表里:CriPT AA 只有 16.3%,无工具 HLE 36.0%,Tau3 Banking 36.1%,AA Omniscience 指数 10.8。安全侧有害请求拒绝\u002F转向率 88.9%。\n\n## 部署生态:「模型是内核」\n\n这次发布的真正主题不是模型,而是 Reflection 说的「开放 Agent 操作系统」:40 余家伙伴接入,覆盖推理、量化、微调、RL、Agent 框架、评估与可观测。Day-0 支持包括 Transformers、vLLM、SGLang、Unsloth、Axolotl 与 NVIDIA Dynamo;AWS、Google Cloud、Cloudflare、Fireworks、Together、OpenRouter、Ollama 等平台提供托管,Dell 与 HPE 负责私有化。\n\n配套的 Mirror CLI 是终端 Agent 框架,Python 前端加 Rust 核心,支持可恢复会话与并行 worktree。训练侧的规模数字:Beam 的 RL 训练共跑了约 13 亿个沙箱,峰值 883,006 个 vCPU,跨 GCP 与 AWS 的 23 个集群。官方还预告了用于领域调优的自蒸馏配方。\n\n## 怎么看\n\n正面说,Apache 2.0 加明确不禁止蒸馏与输出训练,是当前最宽松的一档开源姿态;三精度同发也省掉社区量化的第一波踩坑。保留地说,「开放 Agent 操作系统」目前更多是联盟叙事——40 家伙伴的集成深度需要时间验证,model card 里那几个 10-30 分档的基准也提醒着:长程 Agent 可靠性仍是硬仗。\n\n自托管团队现在可以从权重到生产端点不写一行胶水代码;接下来 30 天,值得盯的是社区微调与第三方评测会不会接住这波开放。\n\n原文:https:\u002F\u002Freflection.ai\u002Fblog\u002Fbeam-is-available\n模型卡:huggingface.co\u002Freflection\u002FBeam-1.0-501B-A23B","https:\u002F\u002Freflection.ai\u002Fblog\u002Fbeam-is-available","9e614db3-0420-45bf-a46d-112af5434c65",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",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},"d11f0044-8aef-487c-bebe-89ce4683a4a3","moe",{"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},"7df74219-9460-4081-8c30-75715c0775f1","en","Beam 1.0 Weights Ship: 501B MoE Open Agent Stack","Reflection AI ships Beam 1.0 weights: 501B\u002F23B sparse MoE, Apache 2.0, three numerics plus a speculative drafter; the card claims ~3x fewer reasoning tokens.","On October 16, Reflection AI put the Beam 1.0 weights on Hugging Face: a 501B-total, 23B-active sparse MoE under Apache 2.0, released in BF16, FP8, and NVFP4 numerics alongside a speculative decoding drafter. Eleven days after the announcement, the startup's first open-weight model has moved from preview to downloadable and self-hostable.\n\n## Architecture: a 1M-context sparse MoE\n\nAccording to the model card, Beam has 501.3B total and 23.4B active parameters; of its 52 layers, 3 are dense and 49 are MoE, with 256 experts. Attention combines gated GQA with sliding windows, the context window is 1M tokens, vocabulary is 152k, the chat template is a modified Harmony format, and the knowledge cutoff is June 30, 2026. It is text-only, positioned for coding, reasoning, and long-horizon agentic workloads.\n\nToken efficiency is listed as the first capability: the card claims Beam reaches solution quality comparable to leading open models using approximately 3x fewer inference-time reasoning tokens. For token-billed agent workloads, that matters more to cost than leaderboard rank.\n\n## Benchmarks: narrow gaps across numerics\n\nAll evaluations ran at temp=0.7, top_p=0.99. Representative BF16 numbers: SWE Bench Verified 81.6%, SWEBench Multilingual 78.3%, Terminal Bench v2.1 80.1%, AIME 2026 97.7%, GPQA Diamond 90.6%, and LiveCodeBench 88.3%. Notably, lower precision barely hurts: on SWE Bench Pro v2-Hard, the NVFP4 build scores 78.6%, above BF16's 76.5%; across the table, gaps between the three numerics mostly stay within about two points.\n\nWeak spots are printed in the same table: CriPT AA at 16.3%, HLE without tools at 36.0%, Tau3 Banking at 36.1%, and an AA Omniscience index of 10.8. On safety, the harmful-request refusal\u002Fpivot rate is 88.9%.\n\n## Deployment: the model is the kernel\n\nThe real theme of the release is what Reflection calls an open operating system for agents: more than 40 partners cover inference, quantization, fine-tuning, RL, agent harnesses, evaluation, and observability. Day-zero support spans Transformers, vLLM, SGLang, Unsloth, Axolotl, and NVIDIA Dynamo; AWS, Google Cloud, Cloudflare, Fireworks, Together, OpenRouter, Ollama, LM Studio, and Microsoft Foundry offer hosting, while Dell and HPE handle on-premises deployments.\n\nThe companion Mirror CLI is a terminal agent harness with a Python frontend over a Rust core, supporting resumable sessions and parallel worktrees. Training-scale numbers made it into the post as well: Beam's RL training ran roughly 1.3 billion sandboxes, peaking at 883,006 vCPUs across 23 clusters on GCP and AWS. Self-distillation recipes are promised in the coming days.\n\n## What to make of it\n\nOn the plus side, Apache 2.0 plus an explicit no-ban stance on distillation and output training is about as permissive as open weights get today, and shipping three numerics at once spares the community its first wave of quantization stumbles. The caveat: an open agent OS is still mostly a coalition narrative — the depth of those 40 integrations will take time to verify, and the 10-30-point benchmarks in the card are a reminder that long-horizon reliability remains the hard part.\n\nTeams that want to self-host can now go from weights to a production endpoint without writing plumbing; observers should watch whether community fine-tunes and third-party evaluations pick up the ball in the next 30 days.\n\nSource: https:\u002F\u002Freflection.ai\u002Fblog\u002Fbeam-is-available\nModel card: https:\u002F\u002Fhuggingface.co\u002Freflection\u002FBeam-1.0-501B-A23B","beam-1-0-weights-open-agent-stack","2026-10-18T21:10:09Z","2026-10-18T21:10:48.667061Z","2026-10-18T21:10:48.667069Z",true,"agent",164,[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},"b44fb160-9027-44cd-b355-a66869004abc","Nemotron 3.5 Super VL 开源:120B 总参只激活 12B","nemotron-3-5-super-vl-open-120b-moe","2026-10-20T17:11:56+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"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":70,"title":71,"news_slug":72,"published_at":73},"cee60a45-f212-4a78-83dd-0226c07b3460","腾讯开源记忆写入模型 Hy-Memory:30B MoE 只激活 3B,四框架平均分超 Opus 5","tencent-hy-memory-30b-memory-writer","2026-10-17T15:14:21+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"ac666c4e-4756-43d8-840a-4eacbf88f804","Meta 开源 816B MoE 模型 Muse Spark Open,跑在两台 Mac Studio 上","meta-muse-spark-open-816b-mac-studio","2026-10-17T03:00:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"1b6ec126-cba1-47da-870b-da19cc523763","Meta 开源 816B MoE 模型 Muse Spark Open:1M 上下文,两台 Mac Studio 跑得动","meta-muse-spark-open-816b-moe-1m-mac-studio","2026-10-15T11:00:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"f3c43720-650b-4054-9349-a1386e06c8fc","Le Chonk 把法国拉回非美\u002F美头部:38 分的 Mistral Large 4","mistral-large-4-le-chonk-intelligence-index-38","2026-10-08T03:30:00+00:00"]