Workflow
人工智能
icon
Search documents
霸屏海外的神秘模型Pony Alpha身份确认 为智谱新一代旗舰模型GLM-5
Zheng Quan Ri Bao Wang· 2026-02-12 04:44
Core Insights - The anonymous model "Pony Alpha" has been revealed to be the test version of Zhipu's flagship model GLM-5 [1] - GLM-5 has achieved the highest scores among current open-source models in several authoritative programming and agent benchmark tests [1] - The model has gained significant attention from global developers, being rated as one of the "strongest anonymous models" during its anonymous testing phase [1] Performance Metrics - GLM-5 scored 77.8 in SWE-bench-Verified and 56.2 in Terminal Bench2.0, marking the highest scores for open-source models [1] - Its performance surpasses that of Gemini 3.0 Pro and is approaching the capabilities of the top proprietary model Claude Opus 4.5 in real programming scenarios [1] - This marks the first time the open-source community has a model that can directly compete with leading players in high-end coding scenarios [1] Technical Features - GLM-5 utilizes a sparse attention mechanism derived from DeepSeek, allowing for low deployment and invocation costs [1] - It offers near-equivalent system engineering capabilities, catering to developers and enterprises that require high-performance AI development assistants while prioritizing data privacy and cost control [1]
智谱在香港大涨至新高 公司上调AI编程订阅套餐价格并将推出新模型
Xin Lang Cai Jing· 2026-02-12 04:34
责任编辑:王永生 智谱在香港早盘最大上涨近34%,股价创上市以来盘中新高。此前公司上调人工智能编码订阅计划 GLM Coding Plan的套餐价格,整体涨幅30%起。截至午盘,智谱涨24.8%,报390港元;本周迄今该股 涨逾90%。另一大模型公司稀宇科技(100 HK)上涨14%。智谱周三表示,周四将正式推出新一代AI模 型GLM-5,旨在处理复杂的编程和智能体任务,并已与Anthropic的Claude Opus系列进行直接对标测 试。 智谱在香港早盘最大上涨近34%,股价创上市以来盘中新高。此前公司上调人工智能编码订阅计划 GLM Coding Plan的套餐价格,整体涨幅30%起。截至午盘,智谱涨24.8%,报390港元;本周迄今该股 涨逾90%。另一大模型公司稀宇科技(100 HK)上涨14%。智谱周三表示,周四将正式推出新一代AI模 型GLM-5,旨在处理复杂的编程和智能体任务,并已与Anthropic的Claude Opus系列进行直接对标测 试。 责任编辑:王永生 ...
深圳着力打造“中国智能体第一城”
21世纪经济报道记者 林典驰 陈思琦 深圳报道 2月12日,21世纪经济报道记者获悉,深圳领先边端智能开放研究院(简称"研究院")在深业上城"城市 云客厅"正式揭牌。 "2026年有望成为'智能体元年'。"深圳市科技创新局局长张林谈到,建设深圳领先边端智能开放研究 院,是抢抓智能体产业发展窗口期、推动边端智能产业高质量发展的关键举措。 三是算力、ICT和软件领域能够提供关键支撑。深圳正加快构建自主可控人工智能软硬件生态,推 动"开源鸿蒙/RISC-V"规模化应用,为边端智能发展提供算力、数据和基础设施支撑。 新成立的研究院,被赋予"技术攻坚、产业升级、生态构建"的核心使命。未来,它将作为深圳边端智能 产业集群的共享知识库与创新策源地,向企业提供源源不断的技术源头供给。 展望2030年,依托研究院及一整套边端智能产业生态,深圳将构建跨领域技术共栈体系,在智能体领域 形成自主可控的核心技术能力,推动一批标志性的智能体项目落地,引领科技创新和产业创新深度融合 新范式,打造具有全球影响力的边端智能产业集群,不断提升深圳在全球产业链中的核心竞争力。 2月10日,深圳市七届人大七次会议举行经济科技专场记者会,披露2025 ...
GLM-5引爆行情!智谱大涨28%
第一财经· 2026-02-12 04:15
Core Viewpoint - The article highlights the successful launch of Zhipu's new model GLM-5, which has received positive market feedback, evidenced by a 28.68% increase in stock price on its first trading day [5]. Group 1: Model Features and Updates - Zhipu's GLM-5 model has enhanced programming and agent capabilities, increasing pre-training data from 23 trillion to 28.5 trillion [6]. - The model introduces a new "Slime" framework to support larger model scales and complex reinforcement learning tasks, along with an asynchronous reinforcement learning algorithm for continuous learning from long-term interactions [6]. - GLM-5 has achieved state-of-the-art (SOTA) performance in coding and agent capabilities, closely matching the user experience of Claude Opus 4.5 in real programming scenarios [6]. Group 2: Applications and Integrations - Typical applications of GLM-5 in agent engineering include end-to-end application development, general agent assistants, and direct output of office documents [7]. - The model can be integrated into the popular open-source AI agent system OpenClaw, allowing users to have a smart intern for various tasks such as web searching and programming [7]. - Zhipu has also launched an AutoGLM version of OpenClaw, enabling seamless integration with Feishu robots [7]. Group 3: Industry Trends - The article notes that multiple model updates are occurring in the industry, focusing on inference efficiency, long context, multimodality, and cost reduction [7]. - Other models released around the same time include Step 3.5 Flash, Qwen3-Coder-Next, and MiniMax-M2.5, all emphasizing similar advancements [7]. - DeepSeek's recent updates have increased context length support to 1 million tokens, a significant improvement from the previous 128,000 tokens [8].
太初元碁等10余家国产AI芯片深度适配MinerU自研模型
Guan Cha Zhe Wang· 2026-02-12 04:14
依托自研的VLM模型,MinerU对PDF及复杂网页的元素捕捉准确率可达99%。无论是精密复杂的数学 公式,还是嵌套繁琐的结构化表格MinerU均能实现精准还原与结构化提取。 据了解,MinerU的核心价值在于其跨行业的普适性与极高的解析精度。对于大模型研发而言,它是高 效的语料生产引擎,能够支撑千万级规模文档向AI-Ready数据的快速转化;对于政企办公及科研领域, 它则是精准的文档解析工具,极大提升了数字化办公的高质量发展。 近期,国内不少主流AI大模型相继推出更新版本,国产AI芯片企业也紧随其后适配新版本大模型。以 太初元碁为例,记者向相关负责人了解到,截至目前其已完成包括DeepSeek、千问、智谱、MinerU、 文心一言等在内的30多个AI大模型的国产算力适配工作,涵盖了Qwen3Dense/MoE系列模型、BAAI Embedding/Reranker系列模型、Qwen-VL、LLaVA等多模态理解系列模型;Stable-Diffusion、FLUX、 Wan系列等多模态生成类模型;GLM、Seed-OSS、文心一言等大语言模型;以及MinerU、DeepSeek- OCR2、Paddle-OC ...
瓴羊携手淘宝,举办首届淘宝电商AI挑战赛
Sou Hu Cai Jing· 2026-02-12 04:12
AI 正在重塑电商经营,也已成为商家重构增长的现实选项。"电商+AI"领域还有哪些未被发掘和满足的 场景与需求,如何通过工具进化帮助商家持续降本增效? 2月 8 日,由浙江省商务厅指导,淘宝开放平台、千问大模型、阿里云、瓴羊联合举办的2026"生意管家 杯"电商AI挑战赛正式开放报名通道(官网地址: https://open.taobao.com/ai2026)。 大赛以"工具进化 生态共赢 商业新生"为主题,提供"350万赛事奖项+千万扶持资源",面向全国淘天电 商人征集经营难题,并向全球AI开发者及优秀企业广发英雄帖,通过"商家出题+线上报名+专家评审 +线下实战"的方式征集优秀电商AI解决方案。 电商AI挑战赛组委会介绍,大赛分为经营提效及商家经营增长两个赛道。前者主要围绕商品管理、消 费者服务、数据洞察等核心场景;后者主要围绕消费者运营、商品运营、营销推广等核心场景。 大赛 的赛题将在 3 月份结合商家出题的结果同步给进入决赛圈的选手,并在 4 月下旬在阿里巴巴总部(杭 州)进行实战对决。届时,由淘天、千问大模型、阿里云、生意管家、瓴羊的产品技术专家、行业运营 专家和知名媒体人等组成的专家评审团,将聚 ...
港股午评:恒指跌0.89%险守27000点、科指跌1.68%,科网股,影视娱乐股调整,AI应用概念股走势分化,电力股拉升,中石化创新高!
Jin Rong Jie· 2026-02-12 04:10
Market Overview - The Hong Kong stock market experienced a collective decline, with the Hang Seng Index down 0.89% to 27,024.06 points, the Hang Seng Tech Index down 1.68% to 5,407.59 points, and the China Enterprises Index down 0.96% to 9,179.06 points [1] - Major technology stocks fell, including Alibaba down 2.12%, Tencent down 2.65%, and Meituan down 4.16% [1] Sector Performance - AI application stocks showed resilience, with Zhiyuan (智谱) surging over 33% after the release of its new flagship model GLM-5, which achieved state-of-the-art performance [2] - New energy and resource stocks were active, with CATL (宁德时代) rising over 4% due to a partnership with Changan Automobile to introduce sodium-ion batteries [2] - Precious metals and mining stocks also saw gains, with Zijin Mining (紫金矿业) up over 4% after a target price upgrade from Citigroup [2] Individual Stock Movements - Capital Airport (首都机场) fell over 3% due to a negative earnings forecast, with expected net losses for 2025 reaching 600 to 760 million yuan [3] - Budweiser APAC (百威亚太) dropped over 4% as its 2025 revenue is projected to decline by 7.7% [3] - Dongfeng Motor Group (东风集团股份) rose over 3% after meeting all conditions for privatization and the spin-off of its Lantu brand [3] Institutional Insights - Analysts noted that the market is in a phase of volume contraction, with a focus on structural opportunities in essential retail and tech hardware sectors [4] - Despite inflows from southbound funds, the overall trading volume in the Hong Kong market remains low, indicating a cautious market sentiment [4] - The Hang Seng Index is currently at a historical low valuation, with a PE ratio of only 12 times, suggesting potential for a rebound in quality stocks that have been oversold [4]
面壁智能发布基于稀疏-线性混合架构SALA训练9B模型
Xin Lang Cai Jing· 2026-02-12 04:07
2月12日,面壁智能正式发布行业首个大规模训练的稀疏-线性注意力混合架构SALA,以及基于该架构 的文本模型MiniCPM-SALA,模型仅有9B参数。据介绍,MiniCPM-SALA不使用投机采样等加速算 法,在云端推理芯片上,当序列长度为256K词元时推理速度高达Qwen3-8B的3.5倍,并支持在云端芯片 和消费级端侧GPU上进行高达一百万词元上下文长度的推理。 (本文来自第一财经) ...
SpaceX收购后遗症?xAI创始团队与核心技术人员接连出走
Jin Shi Shu Ju· 2026-02-12 04:02
周三,马斯克在X上回应了xAI近期多名员工离职一事,称公司最近进行了"重组",以"提升执行速 度",而这"需要与一些人分道扬镳"。 他还补充说,SpaceX的并购可能也是促使部分人离开的原因。卡泽米表示,这一举动并不令他意外, 但他原本预期公司会与特斯拉(TSLA.O)合并。多位专家此前也表示,特斯拉很可能会与xAI和SpaceX进 行整合。 "很多人逐渐意识到,这家公司不会像他们原先想象的那样成为一家高速增长的公司,"卡泽米说, "一家小型AI实验室的成长速度,会比一家市值万亿美元的航天公司快得多。" xAI另一名技术人员沙扬·萨利希安(Shayan Salehian)上周宣布离职,称自己将开始"新的事业"。根据 他的帖子及领英页面显示,萨利希安在马斯克旗下公司工作了7年多,从Twitter到X,再到xAI。 "在X和xAI与埃隆密切共事的过程中,我亲眼看到,当你拒绝接受'不可能'作为答案时,会发生什 么,"他写道, 本周,马斯克旗下xAI的两名联合创始成员宣布离职。自1月以来,xAI还有多名技术人员宣布将离开公 司。此时正值马斯克的SpaceX公司收购了xAI。 xAI联合创始人Tony Wu周二在X上宣 ...
刚刚,MOSS孙天祥创业,直播AI4AI大规模科研
机器之心· 2026-02-12 04:00
在春节来临之际,又一支神秘的华人创业团队,在悄悄的搞大事! 就在刚刚,机器之心发现了一条来自创业公司 Analemma 的首条推文,宣布了一个在过年期间,令人不可思议的直播活动: Analemma 在推文中声称,他们将 面向全球公 开直播部 署端到端 AI 研究系统 FARS 。 他们的目标,是在直播中全自动 完成 100 篇论文 。直播预期将持续一个月,系统将持续自动运行直到生产完成论文输出目标后结束。 在新年之际,Analemma 将彻底揭秘完全由 AI 领衔的科研全流程。史无前例,别开生面的超长直播,开创了 AI 研究系统执行全公开的先河。 FARS:AI 自 进化的科研系 统 都说 AI 已经能够独立完成科研工作了,能够辅助人类,甚至自主进行探索的 AI4S 智能体可以说是层出不穷。一些智能体已经可以独立解决数学界的 Erdos 难 题,就连 陶哲轩都已在使用智能体进行自动化研究 。 但是,哪怕已经近在咫尺,使用 AI 来进行 AI 研究的智能体和相关工作仍是凤毛麟角。在 AI4S 已经相对成熟的当下,AI4AI 最具代表性的工作是去年 DeepMind 推出的 AlphaEvolve ,使用大模型自 ...