Core Insights - DeepSeek is evolving beyond being just a "chatbot" base and is optimizing its large model's energy efficiency through architectural innovations, as evidenced by the recent release of new models and frameworks [1][3] - The competitive landscape is intensifying, with DeepSeek's new models being crucial for maintaining its industry position against major players like Google and OpenAI [1][2] Group 1: Technological Developments - In January 2024, DeepSeek released the Engram architecture, which separates "conditional memory" from "computation," aiming to reduce errors and save computational power [3] - The new model, referred to as MODEL1, is speculated to either be a lightweight model suitable for edge devices or a "long-sequence expert" designed for processing lengthy documents or code [3] - DeepSeek's commitment to cost-effective AI solutions is evident, as it aims to lower token costs, making AI development more accessible to a broader range of developers [4] Group 2: Market Position and Competition - The release of new models is seen as essential for DeepSeek to avoid falling behind competitors like Gemini 3 and GPT-5, which have demonstrated superior performance in various benchmarks [7][8] - Despite DeepSeek's strong position in the open-source community, the company faces pressure from the rapid advancements of closed-source models, which could lead to a loss of developer loyalty [10][11] - The competitive dynamics are shifting, with major internet companies increasing their investments in AI, potentially impacting DeepSeek's market share and the overall landscape for domestic AI companies [13][14] Group 3: Ecosystem and Community Impact - DeepSeek's open-source models, such as DeepSeek-V3 and R1, have gained significant traction, accounting for over half of the open-source token throughput in a short period [8][9] - The company has established a decentralized and pragmatic technical ecosystem, attracting developers interested in self-controlled and private deployments [4][6] - The ongoing developments in the open-source AI community are reshaping the narrative around Chinese AI capabilities, with DeepSeek playing a pivotal role in this transformation [5][6]

2月12日凌晨，智谱正式开源发布新一代基座模型GLM-5，被誉为智能体工程时代最强开源模型。它宣 告大模型编程从"写代码片段"的Vibe Coding时代，正式迈入"完成系统工程"的智能体工程时代。 此前，学界与业界已达成共识，Agentic Engineering（智能体工程）时代已经到来。 智谱GLM-5有望成 为国内第一个真正对齐Anthropic Opus系列的模型厂商。 GLM-5的突破性体现在三大维度： 1. 性能对标顶级闭源模型：在权威编程基准测试中取得开源模型最高分，在真实编程场景中的体验已逼 近当前顶尖的Claude Opus 4.5，尤其在处理复杂系统工程与长周期项目上表现出色。 2. 核心技术全面升级：模型底座参数规模达744B，并创新性地采用异步强化学习框架"Slime"和稀疏注 意力机制，在提升智能的同时大幅降低了部署成本。 3. 重塑开发工作流：随模型同步推出的ZCode工具，允许开发者用自然语言描述需求，模型即可自动拆 解任务，指挥多智能体并发完成编码、调试、预览等全流程，甚至支持用手机远程指挥桌面端Agent， 将生产力无限延伸。 ...

2月12日，智谱通过官微宣布上线并开源GLM-5。目前，GLM-5已完成与华为昇腾、摩尔线程、寒武 纪、昆仑芯、沐曦、燧原、海光等算力平台的深度推理适配。即日起，GLM-5在Hugging Face与 ModelScope平台同步开源，模型权重遵循MIT License。GLM-5已经纳入Max用户套餐，Pro将尽快在5 天内支持，接下来我们将逐步扩大范围。同时，智谱也确认此前在OpenRouter市场上发布的开源模型 Pony，即为GLM-5，"在OpenRouter匿名（Pony）上线后，许多开发者使用GLM-5完成了真正能用、能 玩、能上线的应用"。 ...

大家早上好！这里是今天的早报，每天早上，我都会在这里跟你聊聊昨夜今晨发生了哪些大事儿。 昨夜今晨 【美国1月非农录得13万远超预期，市场削减美联储降息押注】 美东时间周三，美国劳工统计局公布的数据显示，美国1月非农就业人数录得13万人，大幅好于市场预期，为 此前就业增长疲弱的一年画上阶段性句号，也为新一年开局注入更强动能，一定程度缓解了外界对劳动力市场 放缓的担忧，支持美联储维持利率不变的政策路径。 具体数据显示，经季节性调整后，美国1月非农新增就业岗位13万个，远超市场预期的5.5万人，前值(12月份) 被小幅下修至4.8万人。 美国1月失业率录得4.3%，略低于市场预期的4.4%，创2025年8月以来新低。 数据公布后，现货黄金短线跳水近40美元，美元指数短线急升50点，非美货币普遍跳水，美国国债收益率也显 著走高。 【苹果据悉开发新版Siri再次遇挫，多项AI功能或推迟发布】 据媒体援引消息人士报道，苹果筹备已久的升级版Siri计划再次遇到挫折，该项目在最近几周的测试过程中遭 遇问题，可能导致多项备受期待的新功能推迟发布。 知情人士透露，苹果原计划在3月推出的iOS 26.4系统更新中加入这些新功能，但 ...

Core Insights - The anonymous model "Pony Alpha" has been revealed to be the testing version of Zhiyu's GLM-5, generating significant interest in the overseas developer community [1] - GLM-5 achieved the highest scores among current open-source models in several authoritative programming and agent benchmark tests, surpassing Gemini 3.0 Pro [1] - The model's performance is reported to be close to that of the top proprietary model, Claude Opus 4.5, indicating a strong competitive position for open-source solutions in high-end coding scenarios [1] Performance Metrics - GLM-5 scored 77.8 in SWE-bench-Verified and 56.2 in Terminal Bench 2.0, marking it as the highest scoring open-source model [1] - The model's capabilities allow it to compete directly with leading proprietary models, showcasing a significant advancement for the open-source community [1] Technical Features - GLM-5 utilizes a sparse attention mechanism derived from DeepSeek, enabling low deployment and invocation costs while providing robust system engineering capabilities [1] - This model offers an unprecedented open-source solution for developers and enterprises that require high-performance AI development assistants while prioritizing data privacy and cost control [1]

Group 1 - The State Council emphasizes the need to comprehensively promote AI technology innovation, industrial development, and application empowerment to foster new productive forces and drive high-quality development [3] - The State Council aims for a unified national electricity market system to be fully established by 2035, transitioning to unified pricing and joint trading [13] - The National Bureau of Statistics reports that China's CPI rose by 0.2% year-on-year in January, while PPI fell by 1.4%, with the data reflecting a base period adjustment [4][13] Group 2 - The automotive industry in China saw production and sales of 2.45 million and 2.346 million vehicles in January, respectively, with a slight year-on-year increase in production and a decrease in sales [13] - The banking wealth management scale decreased by 100 billion yuan in January, indicating a rebalancing of funds among deposits, wealth management, insurance, and equity assets [13] - The Hong Kong Monetary Authority is actively processing license applications for stablecoin issuers, aiming to position Hong Kong as a global innovation center for digital assets [15] Group 3 - The capital market continues a "zero tolerance" regulatory approach, with numerous penalties issued to listed companies and intermediaries for various violations, reflecting an increase in accountability and comprehensive regulation [9] - The Hong Kong IPO market has seen a rare "zero break" phenomenon, with 22 new stocks listed this year not experiencing any price drops on their first day [9] - The MSCI announced its quarterly index adjustments, including the addition of 37 stocks to the MSCI China Index, which will take effect after the market closes on February 27 [9]

Market Overview - US stock indices experienced slight declines, with the Dow Jones down 0.13%, Nasdaq down 0.16%, and S&P 500 down 0.01%. Notable tech stocks like Google and Microsoft fell over 2%, while Intel rose over 2% [4] - International oil prices rose, with WTI crude oil up 1.66% at $65.02 per barrel and Brent crude up 1.42% at $69.78 per barrel [5] - European stock indices showed mixed results, with Germany's DAX down 0.53%, France's CAC40 down 0.18%, and the UK's FTSE 100 up 1.14% [6] Employment Data - The US non-farm payrolls increased by 130,000 in January, significantly exceeding market expectations of 65,000, marking the largest increase since April 2025. The unemployment rate unexpectedly dropped to 4.3% from the expected 4.4% [4] Government Initiatives - The State Council of China emphasized the importance of AI development and its integration across various industries during a recent study session led by Premier Li Qiang [7] - The State Council issued guidelines to improve the national unified electricity market system, focusing on optimizing resource allocation and ensuring equal participation from various market players [8] Corporate Developments - Bill Gates made a return visit to China, focusing on advancements in public health and technology collaboration [15] - DeepSeek and other AI companies announced significant updates to their models, indicating advancements in AI technology [16][17] - The founder of the retail brand "胖东来" announced retirement plans, which may lead to strategic changes within the company [18] Regulatory Actions - The Beijing Consumer Association held discussions with Huazhu Group regarding unfair terms in their membership agreements, highlighting increased regulatory scrutiny in the hospitality sector [20] - The State Administration for Market Regulation released antitrust guidelines for public utilities, addressing monopolistic behaviors in sectors like water and electricity [9] Financial Announcements - Companies like 新锐股份 and 天汽模 announced plans for significant acquisitions, indicating ongoing consolidation in their respective industries [36] - Several companies reported their 2025 net profits, with notable increases for some, such as 道通科技, which saw a 45.89% rise [39]

Core Viewpoint - The company Zhiyun (2513.HK) confirmed that its new model GLM-5, previously known as "Pony Alpha," has topped the popularity chart on the global model service platform OpenRouter [1] Group 1 - The new model GLM-5 is now available on the chat.z.ai platform [1]

2月11日，智谱正式推出新一代旗舰模型GLM-5，主攻编程与智能体能力，官方称已实现开源领域最优 表现。这是继DeepSeek后，国产AI大模型春节档的又一重要发布。 GLM-5参数规模由上一代的355B扩展至744B，激活参数从32B提升至40B。智谱方面证实，此前在全球 模型服务平台OpenRouter登顶热度榜首的神秘模型"Pony Alpha"即为GLM-5。 架构配置方面，GLM-5构建78层隐藏层，集成256个专家模块，每次激活8个，激活参数约44B，稀疏度 5.9%，上下文窗口最高支持202K token。 编程能力显著提升 新一代旗舰模型GLM-5在内部Claude Code评估集中表现突出。前端、后端及长程任务等编程开发场景 下，该模型较上一代GLM-4.7实现全面超越，平均性能提升逾20%。 GLM-5能够以极少人工干预，自主完成Agentic长程规划与执行、后端重构、深度调试等复杂系统工程 任务。官方称，真实编程环境中的使用体感已逼近Claude Opus 4.5水平。 智谱将GLM-5定位为最新一代旗舰级对话、编程与智能体模型，重点强化其在复杂系统工程与长程 Agent任务中的处理能力 ...

Group 1: Google Chrome and WebMCP Protocol - Google Chrome team has released the WebMCP (Web Model Context Protocol), allowing AI agents to interact directly with website kernels via the navigator.modelContext API, bypassing human user interfaces [1] - WebMCP addresses the high costs and low stability issues of traditional agent screenshot recognition, marking a transition from "visual simulation" to "logical direct connection," referred to as "API in UI" [1] - This standard is being jointly promoted by Google and Microsoft, indicating a potential future division of the internet into UI layers for humans and tool layers for agents, heralding the arrival of the "Agentic UI" era [1] Group 2: Runway's Financing and Model Development - Video generation unicorn Runway has secured $315 million in Series E funding, achieving a valuation of $5.3 billion, with participation from Nvidia, AMD, and Adobe, bringing total funding to $815 million [2] - Runway's Gen-4.5 ranks third in the AI-generated video leaderboard, surpassing models like Google Veo 3 and OpenAI Sora 2 Pro [2] - The new funding will be used to train the next generation of world models, having already launched the general world model GWM-1, which includes variants for explorative environments, dialogue characters, and robotic operations [2] Group 3: xAI Leadership Changes - xAI co-founders Jimmy Ba and Wu Yuhua announced their departures within 48 hours, with 6 out of 12 founding team members having left, including 5 in the past year [3] - Responsibilities of the departing co-founders have been redistributed among other co-founders, and SpaceX's acquisition of xAI has been completed, with an IPO plan set to advance in the coming months [3] - xAI's flagship product Grok has recently exhibited strange behaviors, and the talent loss poses challenges for the upcoming IPO [3] Group 4: DeepSeek's New Model - DeepSeek has quietly launched a new model supporting a 1 million token context window, with knowledge cutoff in May 2025, capable of processing content equivalent to the entire "Three-Body Problem" trilogy [4] - This model remains a pure text model, unable to view images directly but capable of reading text from images and documents, with enhanced Agentic Coding capabilities [4] - The industry trend is shifting from LLM reasoning to Agentic reasoning, as indicated by the latest models from Anthropic and OpenAI, suggesting humans will act as architects directing AI teams in software development [4] Group 5: Zhiyu's GLM-5 Model - Zhiyu has confirmed that the mysterious model "Pony Alpha," which topped the OpenRouter popularity chart, is its new model GLM-5, achieving state-of-the-art performance in coding and agent capabilities [5] - GLM-5's performance in real programming scenarios closely approaches that of Claude Opus 4.5, excelling in complex systems engineering and long-range agent tasks with high tool invocation accuracy [5] Group 6: Ant Group's Omni Model - Ant Group has open-sourced the full-modal model Ming-flash-omni 2.0, the first in the industry to generate voice, environmental sound effects, and music simultaneously on the same audio track [7] - This model excels in visual language understanding, controllable speech generation, and image editing, surpassing capabilities of Gemini 2.5 Pro and Qwen3-Omini-30B-A3B-Instruct [7] - The model employs a unified architecture for deep multi-modal integration, supporting zero-shot voice cloning and fine attribute control, and has been open-sourced on platforms like HuggingFace [7] Group 7: iFlytek's Starfire X2 Model - iFlytek has released the Starfire X2 model, trained on entirely domestic computing power, with overall capabilities matching international top levels, particularly in mathematics, reasoning, and agent tasks [8] - Starfire X2 utilizes a 293 billion MoE sparse architecture, improving inference performance by 50% compared to X1.5, and continues to enhance capabilities in over 130 languages, maintaining industry leadership in key languages for Latin America and ASEAN [8] - Industry applications have been significantly upgraded, with medical capabilities passing authoritative evaluations and educational applications achieving personalized learning through error analysis [8] Group 8: Meituan's LongCat Research Agent - Meituan's LongCat has launched a "deep research" feature, scoring 73.1 in the BrowseComp evaluation, approaching top closed-source models, supporting up to 400 interactions and 256K context [9] - Leveraging Meituan's native capabilities in local life, it creates a real training environment and employs a Rubrics-as-Reward mechanism to address AI hallucination issues, ensuring all recommendations are verifiable [9] - The model utilizes a multi-agent specialized division of labor, automating the entire process from information gathering to research analysis and visualization, capable of generating professional reports for restaurant recommendations and travel planning [9] Group 9: ByteDance's Protenix-v1 Model - ByteDance's Seed team has released Protenix-v1, an open-source model that matches the performance of AlphaFold 3 under strict training data and model size constraints [10] - This model successfully unlocks scaling capabilities during inference, with the prediction success rate for antibody-antigen complexes increasing from 36% with a single seed to 47.68% with 80 seeds [10] - The team has adopted a dual-version strategy, with the standard version aligning with academic benchmarks and the extended version utilizing data from June 2025 for practical drug discovery applications, along with the launch of the PXMeter evaluation toolkit [10]