微软测试全新 AI 记忆工具；曝 DeepSeek 最快明年推进 IPO；苹果第三代 iPhone Duo 折叠手机爆料。
苹果首款折叠屏手机 iPhone Duo 正式交付用户 部分订单预计明年 1 月才能送达
苹果首款折叠屏手机 iPhone Duo 于 2026 年 10 月 23 日正式进入上市交付阶段，首批预购用户开始收到新机，苹果零售店也迎来首发日消费者。由于预购需求强劲，加上生产供应受到限制，上市初期的现货数量相当有限，部分消费者即使在首发日下单，也可能需要等待数月才能拿到设备。
苹果此前表示，iPhone Duo 的预购需求表现异常强劲，公司正在努力尽快将产品交付给消费者。苹果同时确认，全球 Apple Store 门店会提供少量供现场顾客购买的库存，但提醒消费者，这些现货可能很快售罄。对于希望不经预购、直接前往门店购买的用户来说，能否买到新机仍存在较大不确定性。
这次供应紧张，与 iPhone Duo 在预购阶段迅速售罄的情况相吻合。预购于 10 月 16 日开启后，部分配置的预计发货时间很快被推迟。此后，苹果官网显示的新订单交付日期进一步延后，部分订单的预计送达时间已经排到 2027 年 1 月 25 日左右。这意味着，首发需求与现有供应能力之间仍存在明显差距。（来源：cnBeta）
曝 DeepSeek 最快明年推进 IPO，联想系入局 DeepSeek 新一轮融资
据外媒援引知情人士消息，DeepSeek 新一轮融资引入多家产业及财务投资者，其中包括君联资本（Legend Capital）、专注半导体领域的石溪资本及吉利等。与此同时，DeepSeek 已聘请中信证券为其 IPO 提供顾问服务，并于今年 9 月引入新的首席财务官，DeepSeek 最快可能于 2027 年初推进 IPO。
君联资本与联想集团同属联想控股体系。联想集团是联想控股旗下核心产业公司，君联资本则是其重要投资平台。此次君联资本参与 DeepSeek 融资，也意味着联想系资本正式进入 DeepSeek 股东阵营。
从产业协同来看，此次投资进一步拉近了联想系与国内头部大模型厂商之间的关系。联想目前的 AI 业务已经覆盖 AI 终端、企业 AI 和 AI 基础设施多个领域，其中 AI PC、手机、平板等终端需要持续接入大模型能力，企业 AI 业务依赖模型、智能体与行业解决方案结合，服务器及数据中心业务则承接模型训练和推理带来的算力需求。
此前，联想已经围绕 DeepSeek 推进多项产品适配，包括在 AI PC、服务器及企业 AI 解决方案中接入 DeepSeek 模型。此次君联资本进一步参与 DeepSeek 融资，使双方关系从产品和技术适配进一步延伸至资本与生态层面，有望为联想 AI 终端、企业 AI 及 AI 基础设施业务带来更多协同空间。
（来源：新浪科技）
古尔曼爆料苹果第三代 iPhone Duo 折叠手机：配长焦镜头，8 英寸出头内屏
10 月 23 日消息，在第四期 Power On 播客节目中，彭博社的马克 · 古尔曼（Mark Gurman）透露，苹果公司正在规划更大尺寸的iPhone Duo，并考虑在未来机型中加入长焦摄像头。
古尔曼在播客节目中表示，这些规划有望在第三代 iPhone Duo 上实现，该产品预估 2028 年推出，其内屏尺寸可能比现款（7.6 英寸）增加约 0.5 英寸，达到「8 英寸出头」的水平。
古尔曼表示，苹果 iPhone Duo 屏幕尺寸的升级路线，不会推出 8.5 英寸或者 9 英寸的大型折叠屏幕，会采用类似 Pro 系列的渐进式扩屏策略。例如 iPhone 15 Pro Max 从 6.7 英寸增至 iPhone 16 Pro Max 的 6.9 英寸。
影像方面，古尔曼透露长焦「已在苹果的计划清单中」，但尚未确定具体落地时间。他表示，难点不在成本，主要在物理结构，由于折叠机身内部空间有限，长焦模组所需的镜头深度难以容纳。
此外，古尔曼还提到苹果正在测试 iPhone Duo 2 的外屏下 Face ID，并探索加入操作按钮（Action Button）的可能性。若上述升级成真，iPhone Duo 后续机型将同时向「更大内屏、更强影像、更完整交互」方向演进。（来源：IT 之家）
微软低调测试全新 AI 记忆工具 Pantheon
微软正在秘密测试一款面向 Windows 11 的新型人工智能应用，其内部代号为「Project Pantheon」。与此前曾引发广泛讨论的 Recall 功能不同，这款新工具不会自动记录用户屏幕上的所有活动，而是仅保存用户主动选择的内容，因此被外界视为一种更加注重隐私的新型「记忆系统」。
据测试信息显示，Project Pantheon 目前以「Context Vault」的名称出现，同时还涉及「Context IO」和「Projects」等多个内部命名。尽管微软尚未正式公布该项目，但相关应用已经通过未公开的微软商店页面进行测试，并被部分用户发现和体验。
从功能角度来看，Context Vault 更像是一个个人数字记忆库。用户可以主动保存网页、文件、截图、复制的文本内容以及音频和视频资料，并将这些内容整理到不同的收藏集之中。随后，用户能够通过 AI 对这些资料进行搜索、提问和总结，从而快速找回此前保存的信息。
应用主界面包含收藏集、资料库、聊天、时间线、令牌使用情况以及外部连接等多个板块。系统既支持用户手动创建分类，也能够利用人工智能自动生成内容集合。不过测试版本仍处于早期阶段，部分自动生成的分类名称甚至出现了「None」和「null」等明显未完善的情况。（来源：cnBeta）
国产 DRAM 重大突破！长江存储准备量产 9600MT/s 的 LPDDR5X 内存
10 月 23 日消息，据报道，长江存储正在开发速率高达 9600MT/s 的 LPDDR5X 内存，并计划在武汉 Fab 3 工厂实现量产。这是这家 NAND 闪存巨头首次将自研 DRAM 推向高速移动内存赛道，直接对标三星、美光等国际巨头。
长江存储的 DRAM 路线并非临时起意。自 2025 年 10 月起，该公司已在 Fab 2 工厂开展第二代 16Gb LPDDR5 的研发与测试，速率 7500MT/s，月产量约 5000 片晶圆。而此次曝光的 9600MT/s 版本，速率大幅跃升，并首次引入了垂直通道晶体管（VCT）与 4F²单元架构，标志着长江存储正从成熟制程向先进 DRAM 架构快速切换。
4F²架构是这轮技术竞赛的焦点。与传统的 6F²架构相比，4F²能将单个存储单元的面积缩减至四倍特征尺寸的平方，理论上可在相同芯片面积内容纳更多单元，从而提升密度、降低成本。这一路线与长鑫存储此前宣布的 4F² DDR5 计划形成呼应，显示中国存储厂商正集体押注这一新架构。
从 NAND 到 DRAM，长江存储的跨界意味着中国存储产业正从单点突破走向全品类竞争。9600MT/s LPDDR5X 若能如期量产，不仅将补齐国产移动内存的高端空白，更可能在 AI 终端与旗舰手机市场，为国产供应链撕开一道新的口子。（来源：快科技）
华为全球采购认证管理部总裁彭博离职，已履职超 27 年
10 月 23 日下午消息，新浪科技从知情人士处获悉，华为公司董事、全球采购认证管理部总裁彭博已于近日离职。在华为官网的管理层信息一栏中，他的个人履历已被撤下。
据悉，彭博在朋友圈中也发文称，「9952 天，感恩，感激，感谢，继续前行！」证实了从华为离职一事。
彭博退出之后，华为董事会成员中新增了华为终端 BG CEO 何刚。何刚之外，华为常务董事、产品投资评审委员会主任、终端 BG 董事长余承东也在董事会中，出现了终端 BG 两位高管同在董事会的局面。
资料显示，彭博 1976 年出生，于 1999 年加入华为，历任华为 Vodafone 系统部部长、欧洲片区副总裁、运营商 BG 全球销售部总裁及华为西欧地区部总裁等职，主导拓展欧洲市场并突破核心国家业务；2015 年任欧洲区域总裁期间实现手机业务收入增长 5 倍，并在意大利、西班牙等市场取得领先份额；2018 年任华为公司董事会成员，2021 年调任公共关系体系及供应体系工作。（来源：快科技）
短剧也要做防沉迷了！红果短剧上线休息提醒、睡眠提醒机制
近日，红果短剧上线防沉迷「休息提醒」和「睡眠提醒」两项机制。用户看剧超过一定时长，或在深夜时段打开红果短剧时，可能会在推荐页刷到一条特别的视频：熟悉的短剧角色对着镜头，提醒用户放下手机，休息一下。其中「休息提醒」在用户看剧超过一定时长后触发，「睡眠提醒」则面向深夜时段仍在看剧的用户。目前，提醒机制正逐步覆盖更多用户。据了解，红果短剧后续还将上线「时间管理」功能，由用户自行设定休息提醒的间隔和睡眠提醒的时间。
据第三方数据机构统计，2026 年 8 月，在中国月度使用时长排名前 50 的应用中，字节系产品总时长份额达到 41.8%。其中，红果短剧约 125 分钟/天，抖音约 112—120 分钟/天。中国网络视听协会数据显示，2026 年二季度，国内共上线微短剧 23.9 万部，其中 AI 剧 15.3 万部，占比 64%，二季度该占比超过 95%。
而进入 2026 年上半年，国内 AI 剧漫剧用户规模预计迅速突破 6 亿，较此前预估的 2.8 亿实现了超预期的翻倍增长。（来源：快科技）
字节跳动 Seedance 2.5 4K 发布：AI 生成视频跟真人没啥区别了
今天，火山引擎正式发布 Seedance 2.5 4K。通过专项训练，Seedance 2.5 4K 更充分地学习了人物表情、细微动作与画面细节，是 Seedance 2.5 这一代中最强的模型。
官方介绍称，Seedance 2.5 4K 让画面中的纹理、轮廓与细节得到更充分的呈现，整体画面更清晰。再加上 10-bit 色深带来的丰富色彩层次与平滑的渐变过渡，画面质感更能胜任大荧幕。
在影视客户的文戏测试中，4K 版本对人物细微反应的呈现更充分，使得观众更容易读到台词之外的情绪，人物的表演也更有层次、更生动。
而且 Seedance 2.5 4K 支持 Draft（样片）模式，终端用户可先用 480P 样片快速验证分镜、反复打磨镜头，确定满意的效果后再生成 4K 正片。
在该模式下，480P 以更优成本给足创意探索空间，4K 负责呈现极致的视觉效果。目前 4K 版本没有面向普通用户全面放开，采取定向邀请模式，优先给到全球 20 家影视工作室和导演，用来尝试电视剧、电影镜头的前期制作。（来源：快科技）
谷歌推出 Agent Compute Engine，为每个 AI Agent 提供独立计算环境
谷歌公布 Agent Compute Engine（ACE），一项专门为 AI Agent 设计的云计算服务。
与传统云服务器不同，ACE 可以为每个 Agent 分配独立的微型虚拟机，使其在隔离环境中运行代码、操作浏览器和调用外部工具。这套系统还针对 Agent 的运行方式进行了优化。
例如，当 Agent 正在等待模型推理结果或用户确认时，系统可以暂停其计算环境，释放部分资源；当任务需要继续执行时，再快速恢复此前的运行状态。
谷歌表示，在一项涉及 20 万个 Agent 的测试中，相关开源运行时实现了低于 500 毫秒的暂停与恢复延迟。
ACE 目前已经用于支持 Gemini 的部分内部任务。谷歌计划在未来几个月向外部开发者开放预览。（来源：谷歌）
月之暗面发布 Kimi K3.1 Pro：3 万亿级参数，新增 3D 创作能力
月之暗面正式发布 Kimi K3.1 系列首款模型 K3.1 Pro，同时推出高速版本 K3.1 Pro Highspeed。
这是 Kimi 目前能力最强的 3 万亿级参数模型，支持最高 100 万 Token 上下文。高速版在保持相同模型能力的前提下，输出速度提升约 2—3 倍。
新模型重点强化了编程、长程 Agent、3D 创作和金融研究能力。
K3.1 Pro 可以根据文字描述、草图或参考图片，推断物体形状和空间关系，并通过编写 Blender、CAD 等专业工具的脚本生成 3D 模型。
Kimi 网页端和 App 也同步上线了 3D Make 功能，用户可以在线预览、编辑生成的 3D 模型，并将其下载用于 3D 打印。
根据月之暗面公布的数据，K3.1 Pro 在 Terminal-Bench 4.0 上取得 65.7% 的成绩，高于 GPT-6 Astra 的 58.2%，接近 Claude Opus 5.5 的 66.4%。
模型已经上线 Kimi App、Kimi Work、Kimi Code 和 API，完整模型权重计划于 11 月 4 日前开放。
（来源：IT 之家）
复古游戏模拟器 Delta 2.0 上线：苹果 iPhone Duo 变身「掌上双屏游戏机」
科技媒体 9to5Mac 昨日（10 月 22 日）发布博文，报道称复古游戏模拟器 Delta 更新至 2.0 版本，在适配苹果首款折叠iPhone Duo之外，在界面上融入了液态玻璃（Liquid Glass）设计。
在游戏体验方面，Delta 针对 iPhone Duo 的多种使用姿势，设计了专属虚拟手柄皮肤，其中最值得关注的是任天堂 DS 布局。应用可充分利用内屏尺寸，在竖向「书本模式」下跨越折叠铰链，把上半屏作为游戏画面、下半屏作为触控操作区，还原真实 DS 的双屏体验。IT 之家附上相关图片如下：
在横屏分屏场景中，Delta 也做了一个巧妙设计。通常 iPhone Duo 的分屏会将两个应用各限制在约一半屏幕内，但 Delta 把虚拟 DS 的下半屏以自定义键盘视图呈现，让操作区可横跨整个屏幕宽度，甚至覆盖分屏另一侧的视频应用。这意味着用户可以一边看视频，一边在下方完整操作游戏。
在适配 iPhone Duo 之外，Delta 2.0 还兼容支持 Sega Genesis ROM，改进了游戏库导航，并更新音频设置，允许游戏声音与正在播放的音乐或播客混合输出。高级版可通过 App Store 内购订阅解锁，现已面向全球推出。（来源：IT 之家）
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Microsoft testing a new AI memory tool; exposure to DeepSeek to advance IPO at the latest next year; apple 3rd generation iPhone Duo folding phone explosion.
The apple front-line folder phone, iPhone Duo, officially delivered to the user.
An iPhone Duo, an apple front-line folder, officially entered the market delivery phase on October 23, 2026, and the first pre-purchasers started receiving new machines, and the Apple retail store also welcomed first-day consumers. Owing to strong pre-purchase demand, combined with limited production supplies, the availability of cash at the beginning of the market was rather limited, and some consumers may need months to obtain equipment even on the date of issuance.
Apple had previously indicated that the pre-purchase demand for iPhone Duo was exceptionally strong and that the company was working to deliver products to consumers as soon as possible. At the same time, Apple confirmed that the Apple Store Gate would provide a small amount of inventory for on-site customers, but warned consumers that the spot might soon be sold out. There is still considerable uncertainty about the availability of new machines for users who wish to buy them directly without advance purchase.
This supply was tight, consistent with the rapid liquidation of the iPhone Duo at the pre-purchase stage. After the advance purchase was opened on 16 October, the projected shipment time for the partial configuration was delayed very soon. Since then, the delivery date of the new orders shown by Apple Network has been further delayed and some orders have been scheduled for delivery around 25 January 2027. This means that there is still a clear gap between initial demand and existing supply capacity. (Source: cnBeta)
DeepSeek, move on next year at the earliest.
According to sources, the DeepSeek new round of financing has introduced a number of industrial and financial investors, including Legend Capital, Stone Creek, which focuses on the semiconductor sector, and Jili. At the same time, DeepSeek has hired Chinese mail securities to provide consultancy services for his IPO and introduced a new Chief Finance Officer in September this year. DeepSeek is most likely to advance IPO in early 2027.
The League of Capital and Thoughts is part of the joint ownership system. The Consortium is a core industry company that wants to hold shares, and the Kings Union capital is its important investment platform. The participation of the KMC in DeepSeek financing this time also meant that the capital entered the DeepSeek shareholder camp officially.
In terms of industrial synergies, the investment further brought the association closer to the head of LLM manufacturer in the country. Consider that the current AI operations already cover several areas of AI terminal, Enterprise AI and AI infrastructure, including AI PC, mobile phones, tablets, etc., which require continuous access to LLM capabilities, the Enterprise AI Business Dependence Model, AI agent in combination with industry solutions, and server and data centre operations in which the computing needs of model training and reasoning are carried over.
Previously, the Consortium had advanced several product adaptations around DeepSeek, including access to the DeepSeek model in AI PC, servers and enterprise AI solutions. The further involvement of the KMC capital in DeepSeek financing, which further extends the relationship from product and technology adaptation to capital and ecological aspects, is expected to bring more synergy to AI terminal, enterprise AI and AI infrastructure operations.
(Source: New Wave Technology)
IPhone Duo Collapse Phone: Focused shot, 8 in-out Insidescreen
On 23 October, according to news from the fourth Power On podcast, Mark Gurman of Bloomberg announced that Apple was planning a larger iPhone Duo and was considering adding a long-focal camera to the future model.
In the podcast programme, Gulman indicated that these plans were expected to be achieved on the third generation iPhone Duo, a product that was projected to be launched in 2028 and that the inner-screen size could increase by about 0.5 inches over the cash (7.6 inches) to the level of “8 inches out”.
Gulman states that the upgrade route for the apple iPhone Duo screen will not introduce a large folding screen of 8.5 or 9 inches and will employ a progressive expansion strategy similar to the Pro series. For example, iPhone 15 Pro Max increased from 6.7 inches to 6.9 inches of iPhone 16 Pro Max.
In terms of images, Gulman revealed that focal points were "on the list of apples' plans", but no specific landing time has been determined. He stated that the difficulty was not at cost, mainly in the physical structure, and that the depth of shot required for the long focus group was difficult to accommodate because of the limited internal space available for folding.
In addition, Gulman mentioned that apples were testing Face ID under the iPhone Duo 2 outer screen and exploring the possibility of adding to the Action button. If the above-mentioned upgrade is made, the iPhone Duo follow-up will also evolve in the direction of " bigger inner screen, stronger image, more complete interaction". (Source: IT Home)
Microsoft low-key testing new AI memory tool Pantheon
Microsoft is secretly testing a new artificial intelligence application for Windows 11, internally coded "Project Pantheon". Unlike the Recall function, which had previously been the subject of extensive discussion, the new tool would not automatically record all the activities on the user screen, but would simply preserve the content of the user’s choice and would therefore be seen outside as a new “memory system” that is more privacy-oriented.
According to the test information, Project Pantheon currently appears under the name "ContextVault" and involves multiple internal names such as "Context IO" and "Projects". Although Microsoft has not yet officially published the project, the applications have been tested through an undisclosed Microsoft shop page and have been detected and experienced by some users.
From a functional point of view, ContextVault is more like a personal digital memory bank. Users can actively save web pages, files, screenshots, copying text contents and audio and video materials and organize them into different collections. Users are then able to search, question and summarize this information through AI in order to quickly retrieve previously stored information.
The application main interface consists of various segments such as collections, databases, chats, time lines, token usage and external connections. The system supports both the creation of classifications manually by users and the automatic generation of content pools using artificial intelligence. However, the test version is still in its early stages, and some self-generated classification names even appear to be clearly imperfect, such as "None" and "null". (Source: cnBeta)
Nationally produced DRAM Major breakthrough! LPDDR5X memory of the Yangtze storage capacity 9600MT/s
On 23 October, it was reported that the Yangtze storage was developing a LPDDR5X memory at a rate of up to 9600MT/s, and was planning to produce quantities at the Wuhan Fab 3 plant. This is the first time that this NAND flashing giant has moved itself to a high-speed moving memory track, directly against international giants like Tristar and Light.
The DRAM route stored by Yangtze is not intended to be temporary. Since October 2025, the company has been developing and testing the second generation of 16 Gb LPDDR5 at a rate of 7500 MT/s, at a monthly production rate of approximately 5,000 crystals at Fab 2 factory. The release of 9600MT/s, which was exposed this time, made a dramatic leap and introduced for the first time the vertical channel transistor (VCT) and the 4F2 unit structure, marking a rapid transition from a mature to an advanced DRAM structure.
The 4F2 architecture was the focus of the technology competition. Compared to the traditional 6F2 architecture, 4F2 reduces the size of a single storage unit to four times the size of the characteristic square, theoretically allowing more units to be included in the same chip area, thereby increasing density and reducing costs. This route echoes the previously announced 4F2 DDR5 plan for the storage of Changsung, showing that Chinese storage manufacturers are collectively betting on this new structure.
From NAND to DRAM, cross-border storage of Yangtze means that the Chinese storage industry is moving from a single-point breakthrough to full-text competition. 9600MT/s LPDDR5X will not only fill the high-end gaps in the mobile memory of the national production, but may more likely open a new mouth to the national production supply chain at the AI terminal and the flagship mobile phone market. (Source: Fast Technologies)
W.W.'s departure for Managing Director Bloomberg of the Global Procurement Certification Department has been over 27 years
In the afternoon of 23 October, Xinhuatech learned from the source that Bloomberg, Director of the Chinese Company and Managing Director of the Global Procurement Certification Authority, had recently left his post. In the management information section of the Chinese official network, his personal history has been withdrawn.
According to the information received, Bloomberg also wrote in his friends' circles that "9952 days, thanks, thanks, thanks, keep going!" confirmed his departure from China.
After Bloomberg's exit, China added the final BG CEO to the board. In addition to Ho, the President of the Final Board of Directors, the Director of the Product Investment Review Board and the Chairman of the Final Board of Directors of the BG were also on the Board of Directors, where two of their executives were co-located.
According to the data, Bloomberg, born in 1976, joined China in 1999, with China as Minister of the Vodafone System, Vice-President of the European Region, Managing Director of the Global Marketing Department of BG and President of the Western European Region, led the expansion of European markets and breakthroughed core State operations; during his tenure as Managing Director of the European Region in 2015, income from mobile phone operations increased fivefold and gained a leading share in markets such as Italy and Spain; in 2018, he was a member of the Corporate Board of Directors and in 2021, he was transferred to the Public Relations and Supply System. (Source: Fast Technologies)
The short play's about to get lost. Red fruit shorts are on the line.
In recent days, the red fruit shorts have come up on the line and become obsessed with the mechanisms of Rest Alert and Sleep Alert. Users who watch a play for longer than a certain period of time, or who open a red fruit short play in the middle of the night, may draw a special video on the recommendation page: the familiar short play character is pointing to shot, reminding users to put down their cell phones and rest. In this case, the Rest Reminder triggers after the user has seen the show for longer than a certain period of time, while the Sleep Alert targets users who are still watching the show at night. At present, the alert mechanism is gradually reaching more users. It is understood that the red fruit short play follow-up will also be connected to the "time management" feature, with users setting their own intervals for the break alarm and the time for the sleep alarm.
According to the third-party data agency, in August 2026, byte ' s share of the total length of the product reached 41.8 per cent in the application of China ' s 50-year-old ranking. Of which, about 125 minutes/day of red fruit short play and Douyin about 112-120 minutes/day. According to the Chinese Association for Online Audio-Visuals, in the second quarter of 2026, there were 239,000 micro-fellows, of which 153,000, or 64 per cent, were in the country, and more than 95 per cent in the second quarter.
In the first half of 2026, the size of the country's AAI drama users is projected to be rapidly over 600 million, more than double the previous estimate of 280 million. (Source: Fast Technologies)
byte beats Seedance 2.5 4K publish: AI generation is no different from reality.
Today, Volcengine officially publish Seedance 2.5 4K. Through specialized training, Seedance 2.5 4K learns more fully about the face of the person, the fine moves and the details of the picture, which is the strongest model in the generation of Seedance 2.5.
According to the official presentation, the texture, contours and details in the picture are more fully presented and the overall picture is clearer. Added to the rich colour layer and smooth transition of the 10-bit depth, the image sense is better suited to the large screen.
In the video client's essays, the 4K version of the show is more nuanced, making it easier for viewers to read emotions outside of dialogue line, and for character performances to be more structured and lively.
Furthermore, Seedance 2.5 4K supports the Draft (samples) model, which allows end users to quickly validate storyboard with 480P samples, grind shot over and over and over again, and then generate 4K solids.
In this model, 480 Ps are giving creative space at a much higher cost, and 4Ks are responsible for the extreme visual effects. At present, the 4K version is not fully open to ordinary users and is based on a targeted invitation model, giving priority to 20 video studios and directors around the world for use in pre-productions of television dramas and films shot. (Source: Fast Technologies)
Google launched Age Computer Engineering to provide an independent computing environment for each AI Agent
Google publishes Agent Company Engineering (ACE), a cloud computing service specifically designed for AI Agent.
Unlike traditional cloud servers, ACE can assign a stand-alone mini-virtual to each Agent to run codes, operate browsers and call external tools in isolated environments. The system also optimizes the way Agent operates.
For example, when Agent is waiting for model reasoning or user confirmation, the system can suspend its computing environment and release some of its resources; and quickly restore the previous state of operation when the task continues.
Google indicated that in a test involving 200,000 Agents, the associated open source was running with a pause and recovery delay of less than 500 ms.
ACE is currently supporting some of Gemini's internal tasks. Google plans to open the preview to outside developers in the coming months. (Source: Google)
The dark side of the moon publish Kimi K3.1 Pro: 3 trillion-degree parameters, adding 3D creativity
The dark side of the moon is the official publish Kimi K3.1 front model K3.1 Pro, with a high-speed version of K3.1 Pro Highspeed.
This is Kimi's most capable 3 trillion-degree parameter model, supporting a maximum of 1 million Token context. The high-speed version increases output speed approximately 2-3 times while maintaining the same modelling capability.
The new model focuses on strengthening programming, long-range Agent, 3D creative and financial research capabilities.
K3.1 Pro can extrapolate object shapes and spatial relationships from text descriptions, sketches or reference pictures and generate 3D models by preparing scripts of specialized tools such as Blender, CAD, etc.
The Kimi webend and App also synchronized the 3D Make feature, which allows users to preview online, edit the generated 3D model and download it for 3D printing.
According to data published on the dark side of the month, K3.1 Pro achieved 65.7 per cent on Terminal-Bench 4.0, up from 58.2 per cent in GPT-6 Astra, close to 66.4 per cent in Claude Opus 5.5.
The model is on line Kimi App, Kimi Work, Kimi Code and API, and the complete model weighting plan is open by 4 November.
(Source: IT Home)
Retrospect game simulator Delta 2.0 online: iPhone Duo
Technology media 9to5Mac yesterday (22 October) in publish Bo, reported that the retrospect game simulator Delta was updated to version 2.0, incorporating the liquid glass (Liquid Glass) design on the interface, outside of iPhone Duo, which is suitable for apple cash.
In terms of game experience, Delta designed an exclusive virtual handle of skin for the various uses of iPhone Duo, the most important of which is Nintendo DS layout. Apply the full size of the inner screen to cross the folding chain vertically to the "book mode ", using the first half as a game image, the second half as a touch-operating area and restoring the real DS double-screen experience. The following pictures are attached to the IT home:
In the screen segment, Delta also made a brilliant design. Normally, the iPhone Duo segment limits each application to about half of the screen, but Delta presents the second half of the virtual DS as a custom keyboard view, allowing the operating area to span the entire screen width and even over the video application on the other side of the partition. This means that the user can watch the video while fully operating the game below.
In addition to iPhone Duo, Delta 2.0 is compatible with the support of Sega Genesis ROM, improves the game library navigation and updates the audio settings to allow the sound of the game to mix with the music or podcast that is playing. The advanced version, which can be unlocked through inside-App Store, is now available globally. (Source: IT Home)
Auto-fetched and translated by AI Atlas. All rights belong to the original author.