单词学习：![[f2c0ee16-44b8-487d-9f33-1b8c3397a404.mp3]]

新闻播报：![[5c56fdec-431d-44a9-85fb-22dc980252c3.mp3]]

OpenAI发布其最强模型GPT-5 Pro API

简报：

• OpenAI正式推出其迄今最强大的模型GPT-5 Pro API，拥有40万Token的上下文窗口，并支持图像输入。
• 该API的定价设定为每百万Token 15美元。
• 新增了优先处理层，请求处理速度比标准层快40%，并提供服务健康仪表盘以实时监控系统运行状态。

相关链接：

• OpenAI重磅发布GPT-5 Pro API，智能化未来一触即发

⭐ OpenAI Releases Its Most Powerful Model GPT-5 Pro API

Brief:

• OpenAI officially launched its most powerful model to date, GPT-5 Pro API, featuring a 400,000 token context window and support for image input. 🚀
• The API is priced at $15 per million tokens.
• A new priority processing tier has been added, offering 40% faster request processing than the standard tier, along with a service health dashboard for real-time system status monitoring. 📈

Related Links:

• OpenAI's Grand Release of GPT-5 Pro API: An Intelligent Future at Your Fingertips

ChatGPT 转型AI应用平台，集成第三方服务

简报：

• 在OpenAI DevDay2025大会上，ChatGPT宣布引入全新的“应用（Apps）”体系，使用户可在聊天中直接使用第三方服务，如预订酒店、设计和播放音乐。
• 用户首次使用应用需授权，系统也会在相关情境下主动推荐可用应用，如讨论旅行时建议使用Expedia。
• 首批合作伙伴包括Booking.com、Canva、Coursera、Expedia、Figma、Spotify和Zillow等。
• OpenAI发布了Apps SDK（预览版），允许开发者基于开放标准Model Context Protocol (MCP)构建应用，使应用不仅限于ChatGPT平台。
• 未来计划开放应用目录，并推出应用提交审核、变现机制以及对企业和教育版本的支持。

相关链接：

• ChatGPT 迎来重磅升级！应用生态正式上线，开启智能对话新纪元
• OpenAI重磅发布GPT-5 Pro API，智能化未来一触即发
• OpenAI 推出 Sora2 API:颠覆视频生成的新工具

ChatGPT Transforms into an AI Application Platform, Integrating Third-Party Services

Brief:

• At the OpenAI DevDay2025 conference, ChatGPT announced the introduction of a new "Apps" ecosystem, allowing users to directly use third-party services within chats, such as booking hotels, designing, and playing music. 🚀
• Users will need to authorize apps upon first use. The system will also proactively recommend available apps in relevant contexts, such as suggesting Expedia when discussing travel. ✈️
• Initial partners include Booking.com, Canva, Coursera, Expedia, Figma, Spotify, and Zillow, among others.
• OpenAI has released the Apps SDK (preview), allowing developers to build applications based on the open standard Model Context Protocol (MCP), ensuring apps are not limited to the ChatGPT platform.
• Future plans include opening an app directory, launching app submission and review processes, monetization mechanisms, and support for enterprise and educational versions. 💡

Related Links:

• ChatGPT 迎来重磅升级！应用生态正式上线，开启智能对话新纪元
• OpenAI重磅发布GPT-5 Pro API，智能化未来一触即发
• OpenAI 推出 Sora2 API:颠覆视频生成的新工具

蚂蚁集团发布万亿参数开源大模型Ling-1T

简报：

• 蚂蚁集团近日发布其首款旗舰语言模型 Ling-1T，参数量达一万亿，是一款国产开源模型。
• 该模型采用“中训练 + 后训练”的创新训练方法，在超过20T token的高质量语料上进行预训练。
• Ling-1T在推理速度与能力上表现出色，并在代码生成、数学计算和逻辑推理等领域树立了新的标杆。

相关链接：

• Ling-1T 模型页面

Ant Group Releases Trillion-Parameter Open-Source Large Model Ling-1T

Brief:

• Ant Group recently unveiled its first flagship language model, Ling-1T, a domestically developed open-source model boasting a staggering one trillion parameters. 🚀
• The model employs an innovative "mid-training + post-training" methodology, pre-trained on over 20 trillion tokens of high-quality corpus.
• Ling-1T demonstrates outstanding performance in inference speed and capabilities, setting new benchmarks in areas such as code generation, mathematical computation, and logical reasoning. ✨🧠

Related Links:

• Ling-1T Model Page

三星发布700万参数微型模型，特定推理任务表现优于Gemini和Claude

简报：

• 三星SAIL蒙特利尔实验室推出名为“微型递归模型”（TRM）的新型AI架构，参数量仅700万。
• 在数独（Sudoku）和ARC-AGI等复杂的结构化推理任务中，TRM的表现超越了包括Gemini 2.5 Pro和Claude 3.7在内的多个大型语言模型。
• 具体测试数据显示，TRM在ARC-AGI-2上的准确率为8%，优于Gemini 2.5 Pro的4.9%和Claude 3.7的0.7%；同时将Sudoku-Extreme的准确率提升至87.4%。
• 该模型使用不到大型模型0.01%的参数量，证明了小型、针对性模型在处理狭窄、结构化推理任务上的潜力。
• 研究人员强调，TRM并非通用语言模型的替代品，它不适用于开放式、基于文本或多模态的通用任务。

相关链接：

• 700万参数微型模型 TRM，推理能力竟胜过 Gemini2.5Pro 和 Claude3.7

Samsung Releases 7 Million Parameter Micro-Model, Outperforms Gemini and Claude on Specific Reasoning Tasks

Brief:

• Samsung's SAIL Montreal lab has introduced a new AI architecture called "Tiny Recurrent Model" (TRM), featuring only 7 million parameters. 🤖
• TRM has surpassed multiple large language models, including Gemini 2.5 Pro and Claude 3.7, in complex structured reasoning tasks such as Sudoku and ARC-AGI.
• Specific test data shows TRM achieving an 8% accuracy on ARC-AGI-2, outperforming Gemini 2.5 Pro's 4.9% and Claude 3.7's 0.7%. It also boosted Sudoku-Extreme accuracy to 87.4%. 📊
• Using less than 0.01% of the parameters of larger models, TRM demonstrates the potential of small, specialized models in handling narrow, structured reasoning tasks.
• Researchers emphasize that TRM is not a substitute for general-purpose language models and is not suited for open-ended, text-based, or multimodal general tasks. ⚠️

Related Link:

• 7 Million Parameter Micro-Model TRM, Reasoning Capability Surpasses Gemini 2.5 Pro and Claude 3.7

微软Azure为OpenAI启用全球首个英伟达GB300生产集群

简报：

• 微软云计算部门Azure于当地时间10月9日宣布，已推出全球首个生产级的英伟达GB300 NVL72超级计算集群。
• 该集群的首位客户是OpenAI，其工作负载将运行在基于此集群的NDv6 GB300 VM虚拟机上。
• 该集群包含共计4608个英伟达Blackwell Ultra GPU，分布于64组液冷机架系统，并采用Quantum-X800 InfiniBand高速互联。
• 微软表示，未来计划在全球的AI数据中心内部署数十万个英伟达Blackwell Ultra GPU。

相关链接：

• 微软 Azure 推出业界首个生产级英伟达 GB300 NVL72 大型集群，顾客为 OpenAI

Microsoft Azure Activates World's First NVIDIA GB300 Production Cluster for OpenAI

Briefing:

• Microsoft's cloud computing division, Azure, announced on October 9 (local time) that it has launched the world's first production-grade NVIDIA GB300 NVL72 supercomputing cluster. 🚀
• OpenAI is the first customer for this cluster, with its workloads set to run on NDv6 GB300 VM virtual machines powered by the cluster.
• The cluster comprises a total of 4608 NVIDIA Blackwell Ultra GPUs, distributed across 64 liquid-cooled rack systems, utilizing Quantum-X800 InfiniBand for high-speed interconnect.
• Microsoft stated its future plans to deploy hundreds of thousands of NVIDIA Blackwell Ultra GPUs across its AI data centers worldwide. 💡

Related Link:

• Microsoft Azure Launches Industry's First Production-Grade NVIDIA GB300 NVL72 Large Cluster for OpenAI 🔗

英特尔AI芯片路线图公布，将推推理优化GPU并实现年更

简报：

• 英特尔在技术之旅活动上公布其AI执行路线图，宣布其AI GPU产品将与英伟达和AMD一样，实现年度更新节奏。
• 公司计划推出一款专为推理优化的AI GPU，该产品将主打大内存容量与高带宽，以提供企业级的推理性能。
• 未来将推出新一代推理GPU和 "Shores" 级产品线，专为智能体AI和训练负载设计，旨在解决AI芯片的带宽瓶颈问题。
• 更多关于推理GPU的详细信息将在本月中旬的2025 OCP全球峰会上公布。

相关链接：

• 英特尔公布 AI 执行路线图：AI GPU 年更，将发布推理优化 GPU

Intel Unveils AI Chip Roadmap, Plans Inference-Optimized GPUs with Annual Updates 🚀

Brief:

• Intel unveiled its AI execution roadmap at its Technology Tour event, announcing that its AI GPU products will adopt an annual update cadence, similar to NVIDIA and AMD.
• The company plans to launch an AI GPU specifically optimized for inference, which will feature large memory capacity and high bandwidth to deliver enterprise-grade inference performance.
• Future plans include a new generation of inference GPUs and a "Shores" product line, designed for agent AI and training workloads, aiming to address the bandwidth bottleneck issue in AI chips. 💡
• More detailed information about the inference GPU will be announced in mid-this month at the 2025 OCP Global Summit. 🤖

Related Links:

• Intel Unveils AI Execution Roadmap: Annual AI GPU Updates, Launching Inference-Optimized GPUs

Figure AI发布量产型人形机器人，旨在进入家庭和商业领域

简报：

• Figure AI近日发布其第三代人形机器人Figure03，定位为首款专为量产设计的机型，旨在进入家庭和商业实际应用场景。
• 该机器人由Figure自研的“视觉-语言-动作”模型Helix驱动，可通过语音指令完成洗衣、清洁等家务，并能自主爬楼梯和适应复杂环境。
• 硬件方面全面升级：机器人身高1.68米，负重20公斤，续航5小时；视觉系统帧率翻倍、延迟降低75%，并在手掌中内置摄像头；同时自研了高灵敏度触觉传感器。
• Figure AI建立了垂直整合的供应链，计划初始年产能为12,000台；但目前尚未公布定价和上市时间。
• 在《时代》杂志公布的2025年度最佳发明榜单中，Figure03作为家用机器人成功入选。

相关链接：

• Figure AI 推出第三代人形机器人 Figure03:软硬件全面升级，能洗衣、爬楼梯
• 宇树双足人形机器人 Unitree R1 入选《时代》杂志 2025 最佳机器人发明榜单

Figure AI Unveils Mass-Production Humanoid Robot, Aimed at Home and Commercial Sectors

Briefing:

• Figure AI recently unveiled its third-generation humanoid robot, Figure03, positioned as the first model designed specifically for mass production, aiming to enter real-world applications in homes and businesses. 🤖
• The robot is powered by Figure's self-developed 'vision-language-action' model, Helix, enabling it to perform household chores like laundry and cleaning via voice commands, as well as autonomously climb stairs and adapt to complex environments.
• Hardware has received a comprehensive upgrade: the robot stands 1.68 meters tall, can carry 20 kg, and has a 5-hour battery life. Its vision system boasts doubled frame rates and a 75% reduction in latency, with cameras integrated into its palms. Figure also developed high-sensitivity haptic sensors. ✨
• Figure AI has established a vertically integrated supply chain, planning an initial annual production capacity of 12,000 units; however, pricing and availability have not yet been announced.
• Figure03 has been successfully listed as a home robot in TIME Magazine's 'Best Inventions of 2025' list. 🏆

Related Links:

• Figure AI Unveils Third-Gen Humanoid Figure03: Comprehensive Software & Hardware Upgrades, Can Do Laundry & Climb Stairs
• Unitree's Bipedal Humanoid Robot R1 Makes TIME Magazine's 'Best Robot Inventions of 2025' List

Veo 3.1更新，提升角色一致性并支持单图生成故事

简报：

• Veo 3.1版本将生成视频的最长时长提升至1分钟，并支持原生1080p输出。
• 新版本增强了角色一致性，使人物在多段画面中的外观与行为更连贯。
• 加入“多镜头多提示”功能，用户仅提供一张图像，即可自动生成包含多个场景与镜头的短片故事脚本。
• 内置多个电影级调色与风格模板，方便用户快速生成专业级影像效果。

相关链接：

• 生成式视频再进化！Veo 3.1 将带来角色一致性与多场景故事生成功能

Veo 3.1 Update: Improved Character Consistency and Single-Image Story Generation

Brief:

• Veo 3.1 increases the maximum video generation length to 1 minute and supports native 1080p output. 🎬
• The new version enhances character consistency, making characters' appearance and behavior more coherent across multiple shots. ✨
• Introduces a "Multi-Shot Multi-Prompt" feature, allowing users to generate a short story script with multiple scenes and shots automatically from just one image. 📖
• Includes multiple cinematic color grading and style templates, enabling users to quickly produce professional-grade visual effects.

Related Link:

• Generative Video Evolves Again! Veo 3.1 Brings Character Consistency and Multi-Scene Story Generation