[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-caisi-deepseek-v4-pro-8-month-gap-cost-efficient":3,"topics-all":36,"news-related-2fd15c4a-7c8c-42c4-b90b-4ae15396be65":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},"2fd15c4a-7c8c-42c4-b90b-4ae15396be65","DeepSeek V4 Pro独立评测：开源模型逼近前沿，但能力仍差8个月","4月，美国人工智能标准与创新中心（CAISI）对DeepSeek V4 Pro进行了独立评测。结果显示，这款国产开源旗舰模型能力大约落后美国前沿模型8个月，但在成本效率上展现出显著优势。\n\nCAISI的评估涵盖网络安全、软件工程、自然科学、抽象推理和数学五大领域，使用了16个基准测试、35个模型作为参照。结果显示，DeepSeek V4综合能力约等同于GPT-5，落后GPT-5.5约8个月。\n\n不过DeepSeek V4在成本效率上扳回一城：在7个基准测试中，有5个比GPT-5.4 mini更便宜，成本差距从便宜53%到贵41%不等。\n\n在软件工程领域，DeepSeek V4在SWE-Bench上得分74%，仅次于GPT-5.5（81%）和Opus 4.6（79%），领先GPT-5.4 mini的73%。但在网络安全基准CTF-Archive-Diamond上，DeepSeek V4仅得32%，远低于GPT-5.5的71%。\n\n更值得注意的是，DeepSeek官方自评与CAISI实测存在明显差异。DeepSeek自述V4与Opus 4.6和GPT-5.4能力相当，但CAISI的评估表明其实际表现更接近GPT-5水平。这反映出当前AI行业自评与他评之间的方法论分歧。\n\n长远来看，DeepSeek V4 Pro的意义在于开源模型首次逼近美国前沿阵营，这本身就是突破。成本效率与能力之间的权衡也反映了当前模型优化的现实。","https:\u002F\u002Fwww.nist.gov\u002Fnews-events\u002Fnews\u002F2026\u002F05\u002Fcaisi-evaluation-deepseek-v4-pro","97acf9e4-deb3-41bb-8e98-9396e853733d",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"120fa59a-ff6f-4537-9bf5-f818df636a0e","benchmark",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"0a93ec8e-ea39-4693-81de-563ca8c173f7","inference",{"id":18,"name":19,"slug":19,"description":13,"color":13},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"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},"2b081519-844c-4434-bed4-6313404e6e27","en","DeepSeek V4 Pro reviewed: near the frontier, 8 months behind","In April, the US Center for AI Standards and Innovation (CAISI) conducted an independent evaluation of DeepSeek V4 Pro. The results show this domestic open-source flagship model is about 8 months behind US frontier models in capability, but shows significant advantages in cost efficiency.\n\nCAISI's evaluation covered five major domains — cybersecurity, software engineering, natural science, abstract reasoning, and mathematics — using 16 benchmarks and 35 reference models. The results show DeepSeek V4's overall capability is roughly equivalent to GPT-5, about 8 months behind GPT-5.5.\n\nBut DeepSeek V4 pulls back a city in cost efficiency: in 7 benchmarks, 5 are cheaper than GPT-5.4 mini, with cost differences ranging from 53% cheaper to 41% more expensive.\n\nIn software engineering, DeepSeek V4 scores 74% on SWE-Bench, second only to GPT-5.5 (81%) and Opus 4.6 (79%), leading GPT-5.4 mini's 73%. But on the cybersecurity benchmark CTF-Archive-Diamond, DeepSeek V4 only gets 32%, far below GPT-5.5's 71%.\n\nMore notably, there's significant divergence between DeepSeek's official self-assessment and CAISI's real testing. DeepSeek's self-claim is that V4 is on par with Opus 4.6 and GPT-5.4, but CAISI's evaluation shows its actual performance is closer to GPT-5 level. This reflects methodological divergence between self-evaluation and third-party evaluation in the current AI industry.\n\nIn the long run, the significance of DeepSeek V4 Pro is that an open-source model has for the first time approached the US frontier camp, which is itself a breakthrough. The trade-off between cost efficiency and capability also reflects the current reality of model optimization.","caisi-deepseek-v4-pro-8-month-gap-cost-efficient","2026-05-02T07:05:00Z","2026-05-02T07:07:02.369197Z","2026-08-19T02:08:40.142862Z",true,"agent",374,[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},"2d866988-f04f-42e8-85e1-8a49069f9222","MaxProof测试时缩放：MiniMax M3拿下IMO 2025\u002F USAMO 2026双金","minimax-m3-maxproof-imo-2025-usamo-gold","2026-06-14T12:00:00+00:00",{"id":63,"title":64,"news_slug":65,"published_at":66},"3ef2acbe-e09c-4ab6-9104-e8157e729698","开源大模型2026实用评测：DeepSeek V3.2、Llama 4与Qwen3谁更值得部署？","deepseek-llama4-qwen3-spheron-comparison","2026-05-19T13:00:00+00:00",{"id":68,"title":69,"news_slug":70,"published_at":71},"8def771a-d936-4859-930d-02c3011dc55c","LimiX-2 开源：一个模型吃下分类回归插补，表格三榜登顶","limix-2-tabular-foundation-model","2026-09-17T21:09:27+00:00",{"id":73,"title":74,"news_slug":75,"published_at":76},"176b4807-da61-479f-a514-9381cd13319e","SP3O:3 个锚点修复 PPO critic 的平坦化","sp3o-sparse-critic-supervision","2026-09-17T17:10:01+00:00",{"id":78,"title":79,"news_slug":80,"published_at":81},"2e27016d-b90e-45c7-825a-41fd1e435c80","JHU 新研究:组合持续学习机制,百任务记忆留存从 1.2% 提到 34.9%","compose-cl-long-horizon-memorization","2026-09-16T15:10:00+00:00",{"id":83,"title":84,"news_slug":85,"published_at":86},"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"]