[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-upstage-solar-open-2":3,"news-related-5ef8c5da-7221-4e22-bd46-c1ef55c3120d":36},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":23,"news_slug":29,"published_at":30,"created_at":31,"modified_at":32,"is_published":33,"publish_type":34,"image_url":13,"view_count":35},"5ef8c5da-7221-4e22-bd46-c1ef55c3120d","韩国 Solar Open 2：Linear Attention 进 MoE，250B\u002F15B 激活","7 月 23 日韩国 Upstage 全量开源 Solar Open 2（250B-A15B）。每 token 只激活 15B 参数就能在韩文工业基准 Ko-GDPval 拿到 86.8 分（开源 SOTA），量化后只需 2 张 H200 就能部署，把 1M 长上下文压缩到「单机可跑」的工程边界。\n\n架构叫 Hybrid-Attention MoE：48 层里每 4 层一组，1 层 Softmax 加 3 层 Linear 交替。Linear 层靠循环隐状态编码位置顺序，彻底去掉 RoPE，整网只有 12 层保留 KV 缓存，长上下文显存降到同结构纯 Softmax 模型的大约四分之一。MoE 用 320 路由专家加 1 个共享，每 token 激活 8 加 1，配 GQA 64\u002F8 头。\n\n预训练流程也是亮点：由 102B 的 Solar Open 1 通过选择性权重迁移初始化——架构从全 Softmax 切到 Hybrid 后只有 2.3% 的旧权重能复用，其余全部随机初始化，剩下 250B 训练用约 2M GPU 小时（B200）加 12T token 完成。「只复用兼容子矩阵」的跨架构迁移，把冷启动起点抬高，又把训练成本压回「几个 B200 集群跑得完」的范围。\n\n成绩：SWE-Bench Verified 70.4、APEX-Agents 16.6（开源 SOTA）、MCP-Atlas 58.2、LiveCodeBench v6 92.4，对标 DeepSeek-V4-Flash（284B-A13B）和 Mistral Medium 3.5。韩文侧 KMMLU-Pro 78.4、Ko-GDPval 86.8 刷到开源第一，部分超过 Claude Haiku 4.5。\n\n部署友好：vLLM 0.22.0 原生支持，同一服务暴露 Anthropic 兼容和 OpenAI 兼容两种接口，Claude Code 和 Hermes Agent 可零改动接入；附 INT4 和 NVFP4 量化版，显存门槛压到 4 张 H200。许可证为 Upstage Solar License，衍生模型需前缀「Solar」。\n\n判断：Linear attention 在 1M 上下文里的工程价值被验证。KV 缓存只剩纯 Softmax 的四分之一，加上选择性权重迁移把训练成本压回百万级 GPU 小时，「长上下文加高稀疏激活加 2 张 H200」第一次成为开源模型的可达配置。Solar Open 2 不是最强开源模型，但它把「最强开源必须能部署在企业自有机柜上」这条路线坐实了。","https:\u002F\u002Fwww.upstage.ai\u002Fblog\u002Fen\u002Fsolar-open-2","895e85f1-ff1b-444d-979b-62f10a6be5a2",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"6ad31a14-c0da-42df-81fd-564281f768db","agentic-ai",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":18,"name":19,"slug":19,"description":13,"color":13},"7e89b5cc-57db-4f37-bc6d-28919a73931c","model-release",{"id":21,"name":22,"slug":22,"description":13,"color":13},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":28},"9609a3df-583c-4249-a27a-6de23080371c","en","Korea's Solar Open 2: linear attention in MoE, 250B\u002F15B","On July 23, Korea's Upstage fully open-sourced Solar Open 2 (250B-A15B). Activating only 15B parameters per token, it takes 86.8 on the Korean industrial benchmark Ko-GDPval (open-source SOTA) and, after quantization, can be deployed on just 2 H200s — pushing 1M long context into the \"single-machine runnable\" engineering envelope. The architecture is called Hybrid-Attention MoE: across 48 layers, every group of 4 layers, 1 Softmax layer alternates with 3 Linear layers. The Linear layers rely on recurrent hidden states to encode position order, completely removing RoPE; the entire network retains KV cache in only 12 layers, and long-context memory drops to about a quarter of a same-shape pure-Softmax model. MoE uses 320 routing experts plus 1 shared expert, with 8+1 activated per token, paired with GQA 64\u002F8 heads. The pretraining process is also a highlight: it's initialized by selective weight transfer from the 102B Solar Open 1 — after the architecture switches from all-Softmax to Hybrid, only 2.3% of the old weights can be reused, the rest are randomly initialized, and the remaining 250B training is completed with about 2M GPU-hours (B200) plus 12T tokens. \"Reusing only the compatible sub-matrix\" cross-architecture transfer raises the cold-start baseline while pressing the training cost back into the \"few B200 clusters can finish\" range. Scores: SWE-Bench Verified 70.4, APEX-Agents 16.6 (open-source SOTA), MCP-Atlas 58.2, LiveCodeBench v6 92.4, going head-to-head with DeepSeek-V4-Flash (284B-A13B) and Mistral Medium 3.5. On the Korean side, KMMLU-Pro 78.4 and Ko-GDPval 86.8 hit open-source #1, partially exceeding Claude Haiku 4.5. Deployment-friendly: vLLM 0.22.0 has native support, the same service exposes both Anthropic-compatible and OpenAI-compatible interfaces, so Claude Code and Hermes Agent can be plugged in with zero changes; INT4 and NVFP4 quantized versions are attached, pushing the memory floor to 4 H200s. License is Upstage Solar License; derivative models need the \"Solar\" prefix. Judgment: linear attention's engineering value in 1M context is verified. KV cache drops to a quarter of pure Softmax, plus selective weight transfer compresses training cost back to the million-GPU-hour range — \"long context + high sparse activation + 2 H200s\" becomes an achievable configuration for open-source models for the first time. Solar Open 2 isn't the strongest open-source model, but it cements the line that \"the strongest open source must be deployable in an enterprise's own rack\".","upstage-solar-open-2","2026-07-23T03:30:00Z","2026-07-23T06:11:49.197347Z","2026-08-19T02:08:40.142862Z",true,"agent",102,{"items":37},[38,43,48,53,58,63],{"id":39,"title":40,"news_slug":41,"published_at":42},"f26ace13-9c96-47ea-a528-b6682a22aa1e","Apodex 1.1 把推理搬进真实执行:PIVOT-RL 定位关键决策点,35B mini 开源","apodex-1-1-agentic-execution-pivot-rl","2026-08-25T14:30:00+00:00",{"id":44,"title":45,"news_slug":46,"published_at":47},"b4754043-6b19-499f-8459-f8fc786f4d80","Pokee-Isaac 28B 把 10M 上下文塞进客户边界:28B 参数在 RULER 10M 上 93.3%","pokee-isaac-28b-10m-context","2026-08-20T14:00:00+00:00",{"id":49,"title":50,"news_slug":51,"published_at":52},"36055e5f-136f-497d-8763-3ed6609f59ff","Meta Muse Glimmer 30B 本地落地:Apache 2.0 的开源智能体,把 Agent 装进 24GB 显存","meta-muse-glimmer-30b-local-agent-apache2-r2","2026-08-19T03:00:00+00:00",{"id":54,"title":55,"news_slug":56,"published_at":57},"5878a668-282c-4b88-b2b8-7eef40b7938c","LFM2.5-2.6B：2.5GB 内存跑本机 Agent 220 tok\u002Fs","lfm2-5-2-6b-on-device-agent","2026-08-11T00:00:00+00:00",{"id":59,"title":60,"news_slug":61,"published_at":62},"79c1684f-f61d-4799-b3d0-6450c4ad10e8","Muse Glimmer:Meta 把 30B 「常驻本地的智能体」开源,把 Agent 拉到笔记本里 7×24 跑","muse-glimmer-30b-open-agentic-local","2026-08-10T00:00:00+00:00",{"id":64,"title":65,"news_slug":66,"published_at":67},"c94766df-827e-4e4e-a006-b6639ec76722","DeepSeek V4-Flash-0731 转正观察:权重不动,后训练把 Agent 分数打到 V4-Pro 之上","deepseek-v4-flash-0731-agent-benchmark-official-aug2026","2026-08-01T02:00:00+00:00"]