[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-10m-token-context-race-cools-1m-effective":3,"topics-all":36,"news-related-80093416-3107-46bf-b871-7a9c523d0ba3":55},{"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},"80093416-3107-46bf-b871-7a9c523d0ba3","上下文竞赛降温：10M token 质量真相","上下文竞赛降温：10M token 质量真相\n\n4月5日 Meta 发布 Llama 4 Scout，宣称支持 1000 万 token 上下文——没有任何开源模型接近这个数字。但拨开营销话术，实际情况要复杂得多。\n\nLongBench v2、RULER 等标准化评测显示，当前模型的有效检索范围大多落在 128K 到 1M token 之间。一旦超过这个范围，模型在关键信息定位、推理连贯性上的表现开始显著下滑。10M 是技术上限，1M 才是当前算法和硬件约束下的有效甜点。\n\n与冲击理论上限的路线不同，Qwen3.6-Plus 和 Claude Opus 4.7 选择了在 1M 上下文内做到最优。阿里 4 月 29 日发布的 Qwen3.6-Plus 主打 Agent 编程场景，1M token 足以覆盖完整代码仓库分析，且编程质量更稳定—— Llama 4 Scout 赌的是宽，Qwen3.6-Plus 赌的是深，两者定位根本不同。\n\n上下文长度竞赛正在经历一次理性回归。10M token 证明了工程上可以实现，为超长文档处理奠定了基础；但在当前注意力机制和硬件的约束下，1M token 才是生产环境的现实选择。对于正在选型的工程师来说，关键是问自己：你要处理的是一整座图书馆，还是一层楼的藏书？前者值得押注 10M token 的未来，后者现在选 1M token 模型就够了。","https:\u002F\u002Ftokenmix.ai\u002Fblog\u002Fllm-context-window-explained","618e947b-dd18-4090-a509-7f4d23953dbd",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"0a93ec8e-ea39-4693-81de-563ca8c173f7","inference",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},"b1853a5a-d940-42b7-94f9-0488ee3f2cf7","new-model",{"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":13},"74a4dca7-20a4-4532-92b8-5b0aa28a1e45","en","The Context Window Race Cools: The Truth About 10M Token Quality","On April 5, Meta released Llama 4 Scout, claiming support for 10 million tokens of context — no open-source model comes close to this number. But stripping away the marketing, the actual situation is more complex.\n\nStandardized evaluations like LongBench v2 and RULER show that current models' effective retrieval range mostly falls between 128K and 1M tokens. Once this range is exceeded, model performance in key information location and reasoning coherence starts to drop significantly. 10M is the technical upper limit, 1M is the effective sweet spot under current algorithm and hardware constraints.\n\nUnlike the route of pushing theoretical limits, Qwen3.6-Plus and Claude Opus 4.7 chose to be optimal within 1M context. Alibaba's Qwen3.6-Plus, released April 29, focuses on Agent coding scenarios, with 1M tokens sufficient for full code repository analysis, and more stable coding quality — Llama 4 Scout bets on width, Qwen3.6-Plus bets on depth, the two are fundamentally different positioning.\n\nThe context-length race is experiencing a rational return. 10M tokens proves it's possible engineering-wise, laying the foundation for ultra-long document processing; but under current attention mechanism and hardware constraints, 1M tokens is the realistic choice for production environments. For engineers choosing models, the key is to ask yourself: do you want to process an entire library, or just one floor's worth of books? The former is worth betting on the 10M token future, the latter can just pick a 1M token model now.","10m-token-context-race-cools-1m-effective","2026-04-30T02:08:00Z","2026-04-30T10:08:56.475379Z","2026-08-19T02:08:40.142862Z",true,"agent",188,[37,46],{"slug":38,"tag_slug":38,"title_zh":39,"title_en":40,"intro_zh":41,"intro_en":42,"id":43,"is_active":33,"created_at":44,"modified_at":45},"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":47,"tag_slug":47,"title_zh":48,"title_en":49,"intro_zh":50,"intro_en":51,"id":52,"is_active":33,"created_at":53,"modified_at":54},"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":56},[57,62,67,72,77,82],{"id":58,"title":59,"news_slug":60,"published_at":61},"559df680-f13e-48f0-9263-73116aaff411","上交大DLite免历史投机解码:Agent提速2.74倍","dlite-history-free-speculative-decoding","2026-10-19T17:07:52+00:00",{"id":63,"title":64,"news_slug":65,"published_at":66},"ab1e6bdf-9433-41f0-be4d-cc0f78523ac2","SonicBoom v0.0.3:一个 .so 干掉手工调优","sonicboom-v003-mlir-compiler","2026-10-19T09:03:05+00:00",{"id":68,"title":69,"news_slug":70,"published_at":71},"ad9f5296-75bf-434c-a527-dc2703eb8c6a","Janas 引擎:80B 模型塞进 32GB 内存笔记本","janas-ssd-streaming-moe-inference","2026-10-14T13:12:19+00:00",{"id":73,"title":74,"news_slug":75,"published_at":76},"9a77e0d7-8164-494e-984d-54cb57a7a0dd","Magnitude 开源:本地 Agent 专用推理引擎","magnitude-self-tuning-local-agent-inference-engine","2026-10-01T17:08:35+00:00",{"id":78,"title":79,"news_slug":80,"published_at":81},"c99751d5-418e-49d5-99d3-e43b84c80ec7","IBM与NASA开源月球基础模型:Lunar Foundation Model","nasa-ibm-lunar-foundation-model-sombench","2026-09-19T09:30:00+00:00",{"id":83,"title":84,"news_slug":85,"published_at":86},"d056f67b-7e0d-4e44-8d39-e31ea50deeae","Bonsai 2 27B 三元压缩:Qwen3.8 压到 5.9 GB,benchmark 留存 98.2%","bonsai-2-27b-ternary-qwen3-8-compression","2026-09-17T15:47:00+00:00"]