AI NEWSLatest AI NewsArticleByteDance Aims for 5 Trillion Parameters: Dou Bao’s Intelligence is Expected to Reach Its Peak, at the Cost of Million-Level GPU Computing PowerPublished in Latest AI NewsTime :Aug 7, 2026Read :4minuteParametric scale of domestic large models is continuing to rise beyond 10 trillion. Qwen3.8 Max and Kimi K3 have already reached 2.4 trillion and 2.8 trillion parameters, respectively, and the next step could be 5 trillion. This time, ByteDance may take the lead. LatePost reported that ByteDance is discussing the training of a large model with more than 5 trillion parameters, which would become the largest known model in China. However, the plan is still in its early stages, and it may not ultimately be released.What is the scale of 5 trillion parameters? It is considered to be at the level of GPT-5.6 or Opus 5. Generally, the larger the number of parameters, the stronger the performance of a large model. OpenAI and Anthropic have maintained their leading positions by having higher parameter counts—reports suggest that both companies are training or have completed preparations for 10 trillion parameter large models. However, extremely high parameter counts also mean extremely high computing power investments. Based on H100, training a 5 trillion parameter model would require 100,000 H100s running for 347 days, or one million graphics cards running for 35 days. A million graphics cards have entered the GW-level computing power scale, with extremely high barriers. ByteDance will not be able to invest such a huge amount of computing power all at once. A more reasonable approach would be to train with 200,000 to 300,000 graphics cards for 3 to 4 months, then add preparation and post-training phases, which would take at least half a year, or even a year, before it could be completed.In terms of computing power reserves, OpenAI currently has the most abundant resources. SemiAnalysis estimates that it has about 2 million graphics cards; Anthropic has about 620,000; DeepSeek only has around 60,000, many of which are relatively outdated H20 and H800. With its financial strength, ByteDance can easily feed a 5 trillion parameter large model. With it, Dou Bao’s capabilities could reach the level of GPT-5.6 or Opus 5 without doubt, truly maximizing its intelligence, no longer needing to be mocked by netizens. The only remaining uncertainty is commercial returns—can this astronomical investment bring equivalent returns in practical applications?

AI NEWSLatest AI NewsArticleTikTok Makes a Key Move in AI, New Primary Department EmergesPublished in Latest AI NewsTime :Aug 12, 2026Read :3minuteRecently, ByteDance has once again made organizational adjustments in the field of artificial intelligence, officially establishing a new AI first-level department, mainly focusing on the construction of data and security for large models. The department is led by Wang Yinglei (Adam Wang) and operates in parallel with core departments such as Seed, Flow, and Douyin. This is another important move in ByteDance’s AI strategy, following the layout of Seed and Flow at the end of 2023.It is understood that this newly established department did not emerge out of nowhere. One of its predecessors was the Global Data team, established in 2023. The team originally had around 100 members and primarily provided underlying data support for international business segments such as Dola and TikTok. As the large model business deepened, it gradually took on the responsibility of data procurement and quality management for Seed model training.After internal integration, the upgraded team’s business scope will comprehensively cover foundational models and application businesses, committed to providing high-quality cross-modal data services for all models under ByteDance. Specific work includes standard setting, data procurement, synthesis and cleaning, as well as quality assessment, providing a solid data foundation for the efficient and stable operation of large models.

AI NEWSLatest AI NewsArticleTencent Releases Second Quarter Report: Dual Breakthroughs in Models and AI Products, Continued Heavy Investment in AIPublished in Latest AI NewsTime :Aug 12, 2026Read :2minuteOn August 12, Tencent Holdings (00700.HK) released its second-quarter report: Hy3 of the Hunyuan series completed a rapid iteration from preview to the official version, with the number of calls exceeding 68 times that of the previous generation model within one week of its launch; a larger parameter version, Hy4, is also planned for release in the near future; WorkBuddy and CodeBuddy continue to lead the domestic AI-native intelligent agent market, with data from Analysys showing that WorkBuddy’s PC end traffic exceeded 20 million visits in June; the WeChat AI assistant “Xiaowei” is currently in a gradual testing phase.In the second quarter, Tencent’s gaming revenue increased by 11%, with local gaming revenue growing by 17% and international gaming revenue increasing by 4% on a fixed exchange rate basis; marketing services revenue increased by 22%; financial technology and enterprise services revenue increased by 9%, with cloud services benefiting from AI-driven growth, expansion of international operations, and increased demand for general cloud services.

AI NEWSLatest AI NewsArticleMusk Makes Shocking Remarks: 99% of SpaceX’s Value Will Come from AI Rockets, and Space Exploration Has Become a Side BusinessPublished in Latest AI NewsTime :Aug 12, 2026Read :4minuteSpaceX has released a video of Elon Musk’s latest speech to employees. While reviewing the company’s development journey, he systematically outlined the three major directions of Starship, Starlink, and AI for the first time, with AI becoming the absolute focus. Musk clearly stated that SpaceX not only needs to succeed in the field of AI hardware but also achieve breakthroughs in AI software, and he made an astonishing prediction: “The future essentially belongs to AI and robots; the scale of digital intelligence may ultimately far exceed that of biological intelligence.“Musk further provided quantitative goals: “AI revenue may exceed the combined income from rockets, spacecraft, and Starlink satellite internet services this September,” and will significantly lead in the fourth quarter. He even asserted that AI will account for about 99% of SpaceX’s company value within the next five years. In his view, humanity needs to build an AI that cares about humans and promotes human development, which is the fundamental reason why SpaceX must personally participate in the AI competition.The Ambition of 10GW Computing Power, AI Revenue Approaching $2.6 BillionThe basis for this assessment is SpaceX’s rapidly expanding AI computing landscape. Musk revealed that the company aims to achieve 10 gigawatts of AI computing power by the end of next year, more than seven times the current capacity of about 1.4 gigawatts. These computing resources are currently mainly provided to technology companies and artificial intelligence laboratories. He estimates that if the goal is achieved on schedule, it could bring annual revenues of $300 billion to $500 billion, “a huge number.“SpaceX’s second-quarter financial report confirmed the explosive growth of the AI business: quarterly revenue was approximately $7.8 billion, a 92% increase year-on-year, with AI business revenue reaching nearly $2.6 billion, a sharp rise of 247%; connection business revenue from Starlink satellite internet was approximately $4.3 billion, up 66% year-on-year; traditional aerospace business revenue was approximately $962 million, now the smallest of the three segments. At the same time, the company’s net loss narrowed to $541 million, and adjusted EBITDA reached $3.5 billion, up 191% year-on-year. A company known for rockets is rapidly transforming into an AI computing giant, and Musk’s five-year vision of “99% of the value coming from AI” is gradually being realized on the financial statements.

ENLatest AI NewsTencent AI Advances Across the Board: Hy4 to be Launched Soon, WorkBuddy Continues to Lead, and Weixiao Expands Gray TestingTencent’s second-quarter financial report shows that AI has become the core focus: the Hunyuan large model is rapidly iterating, from Hy3 preview to the official version, and plans to launch Hy4 with even larger parameters; the office intelligent agent has initially established an advantage, with multiple products making progress; the Weixiao AI assistant for WeChat was launched in June and is now expanding gray testing. The fundamentals remain stable, with revenue reaching 204.79 billion yuan, showing growth year-on-year.6 hours ago11.5KMusk Makes Shocking Remarks: 99% of SpaceX’s Value Will Come from AI Rockets, and Space Exploration Has Become a Side BusinessIn an internal SpaceX speech, Musk highlighted Starship, Starlink, and AI as three key directions, with AI as the absolute core. He stressed that SpaceX must achieve breakthroughs in both AI hardware and software, asserting that “the future essentially belongs to AI and robots, and the scale of digital intelligence will far surpass biological intelligence.” He also proposed quantitative goals for AI business…..7 hours ago9.7KTencent Releases Second Quarter Report: Dual Breakthroughs in Models and AI Products, Continued Heavy Investment in AITencent Q2: Hunyuan model Hy3 weekly calls surged 68x, Hy4 upcoming. AI office tool WorkBuddy had over 20M monthly visits, WeChat AI ‘Xiaowei’ in grey testing. Game revenue +11% YoY, local games +17%…..7 hours ago9.0KxAI Launches Grok Bot: Provide AI Agents with a Dedicated Cloud Computer, Not a Chat ToolxAI released a beta Grok Bot—a team of persistent AI agents with cloud PCs. They log into your tools, autonomously run multi-step tasks without supervision, and mimic human workers: tasks proceed in the cloud even if you close your laptop, alerting only for approvals. Manage via desktop…..7 hours ago10.2KTikTok Makes a Key Move in AI, New Primary Department EmergesTikTok has recently established a new primary AI department, focusing on large model data and security construction, led by Wang Yinglei, and is parallel to core businesses such as Seed, Flow, and Douyin. The department was previously the Global Data team formed in 2023, marking another key move by TikTok in the AI field, following Seed and Flow.7 hours ago9.0KFormer Tongyi Qianwen Leader Lin Junyang Officially Launches Startup Pragmatik Labs Secures Hundreds of Millions in Funding with a Valuation of 2 BillionLin Junyang, former leader of Alibaba’s Tongyi Qianwen, established the AI lab Pragmatik Labs (Pragmatic Technology) in Shanghai, focusing on developing general intelligent agents that bridge the digital and physical worlds, proposing the core concept of “Agents for Digital and Physical Worlds.” The company secured its first round of funding through joint investments from GaoRong Capital and Sequoia China.8 hours ago10.2KOpenAI quietly launches paid quota reset feature for ChatGPT Plus, 8 dollars can immediately restore usage quotaChatGPT Plus has launched a paid quota reset feature. Users can pay 8 dollars to immediately restore 100% of their quota after it is exhausted, avoiding about 5 hours of cooling time, or choose to wait for automatic recovery without paying. OpenAI said this move is to improve usage flexibility.8 hours ago8.7KNo need to wait when the quota is used up! ChatGPT Plus quietly launches an $8 reset buttonOpenAI is conducting a gradual test of the paid reset feature: ChatGPT Plus users can pay $8 to immediately restore their quota after it is exhausted, skipping the approximately 5-hour waiting period. A spokesperson confirmed that they are exploring options for purchasing additional quota. Previously, they introduced cumulative resets for Codex, and this complete reset option aims to improve usability.8 hours ago9.4KOpenAI Launches ChatGPT Desktop Version for Linux: Completing the Last Piece of the Puzzle, Competing Head-on with Anthropic for DevelopersOpenAI releases a Linux preview of ChatGPT desktop app, integrating ChatGPT, Work, and Codex, covering Ubuntu, Debian, and Fedora. DEB and RPM packages support x64/ARM64, auto-add repositories for system updates. This completes the desktop suite after Mac and Windows…..8 hours ago9.4KSaber Denies Using AI to Replace Game Scriptwriters, Former Chief Screenwriter Immediately Retorts: You Can’t Have It Both WaysSaber Interactive’s new game ‘Rideshare Simulator’ faces controversy after former lead writer Stella Sacco alleges she was replaced by ChatGPT mid-development. All passenger voice acting was also AI-generated. Sacco questions the lack of disclosure on the Steam page regarding AI’s role in creation…..8 hours ago10.2KGoogle Gemma 4 Global Downloads Surge Past 300 Million, the New Era of AI Has Fully Erupted!Google’s Gemma4 series models have surpassed 300 million global downloads, a milestone announced during Google I/O S

CosyVoice Studio – 阿里推出的一站式AI语音生产力平台AI工具4天前更新AI小集02CosyVoice Studio是什么CosyVoice Studio是阿里巴巴推出的国内首个一站式AI语音生产力平台，基于自研Qwen-Audio模型打造。平台含三大模块，CosyFlow实现语音转写叠加语义理解，自动过滤口语词并生成结构化文本与智能摘要；CosyAgent支持自然语言创建具备企业知识与实时语音交互能力的智能体，面向客服等场景；CosyCreative内置上千种音色并支持声音复刻，可基于文档生成播客与有声书。CosyVoice Studio的主要功能CosyFlow 语音记录：在语音转文字基础上叠加语义理解与文本生成，自动过滤”嗯””那个”等口语填充词，将零散口述整理为邮件、工作汇报、会议纪要等结构化文本。随记功能：面向会议、访谈、课堂等场景，支持实时录制与离线上传，基于声纹区分不同发言人，录音结束后自动生成结构化章节与智能摘要。CosyAgent 语音智能体：通过自然语言快速创建具备企业知识、工具调用与实时语音交互能力的智能体，兼容 prompt 与 workflow 配置，适用于智能客服、电话营销等场景。CosyCreative 音频创作：内置上千种音色并支持声音复刻，用户上传文档、网页链接或输入文字即可生成对话播客或多角色有声书，覆盖音频内容创作全链路。如何使用CosyVoice Studio用户可在iOS、Android、Mac和Windows下载 CosyVoice。CosyAgent和CosyCreative目前以白名单邀请制面向企业用户测试。CosyVoice Studio的核心优势全链路一站式平台：国内首个覆盖”语音记录—智能交互—音频创作”三大环节的 AI 语音生产力平台，个人与企业场景兼顾。自研模型底座：基于阿里巴巴自研 Qwen-Audio 语音模型，中文语音理解、识别与生成能力突出，技术自主可控。语义级语音转写：CosyFlow 在转写基础上叠加语义理解与文本生成，自动过滤口语填充词，将零散口述直接输出为邮件、会议纪要等结构化文本。声纹智能归档：随记功能支持声纹区分发言人，录音结束后自动生成结构化章节与智能摘要，适配会议、访谈、课堂等长语音场景。零代码语音 Agent：CosyAgent 支持自然语言快速创建具备企业知识、工具调用与实时语音交互能力的智能体，兼容 prompt 与 workflow 配置，部署门槛低。多角色音频创作：CosyCreative 内置上千种音色并支持声音复刻，上传文档或输入文字即可生成对话播客与多角色有声书，覆盖音频内容创作全链路。CosyVoice Studio的同类竞品对比维度CosyVoice Studio讯飞听见ElevenLabs定位一站式 AI 语音生产力平台语音转写与会议助手AI 语音合成与克隆平台核心能力转写、Agent、创作全链路覆盖转写、翻译、会议纪要TTS、Voice Clone、API 服务语音转写语义理解+结构化输出，过滤口语词中文转写精准，支持实时与离线不支持语音智能体自然语言创建实时对话 Agent无无音频创作上传文档生成多角色播客/有声书不支持支持多语言 TTS 与声音克隆中文优化强，基于 Qwen-Audio 自研模型强，深耕中文语音多年中等部署方式全平台 App + 企业白名单App + SaaS 网页API + Web 工作台CosyVoice Studio的应用场景会议记录与纪要生成：用 CosyFlow 实时录制会议，自动区分发言人并输出结构化章节与智能摘要，减少人工整理成本。智能客服与电话营销：通过 CosyAgent 自然语言创建具备企业知识与工具调用能力的语音智能体，7×24 小时处理呼入呼出业务。播客与有声书制作：用 CosyCreative 上传文档或输入文字，选择多角色音色自动生成对话播客与长篇有声书，覆盖音频内容创作全链路。访谈与课堂内容归档：借助随记功能离线上传长语音，基于声纹识别区分讲师与学生，自动归档为带时间戳的结构化文本。口述内容结构化整理：用户通过语音口述零散想法，系统自动过滤口语填充词并润色为邮件、工作汇报等正式格式文本。# AI工具©版权声明本站文章版权归AI工具集所有，未经允许禁止任何形式的转载。上一篇SALMONN-2 - 清华等开源的通用音频大语言模型下一篇Warp Agent CLI - Warp 推出的独立命令行编码工具相关文章FilMaster – 港大联合快手、微软、清华推出的AI电影制作系统AI小集2Project AVA – 雷蛇推出的 AI 游戏伴侣AI小集3Toolhouse – AI Agent开发与管理平台，一键部署生产环境AI小集2胃之书 – AI智能饮食记录应用AI小集3一点妙笔 – AI公文写作工具，一键生成合规材料AI小集5秒画趣拍APP – 商汤科技推出的AI创意写真工具AI小集4暂无评论再想想发表评论暂无评论…热门工具Loomy即梦AISekoAiPPT秘塔AI搜索妙呀堆友Agent美图设计室绘蛙AIupdream办公小浣熊秒哒最新收录知漫剧QwenPawRenoiseLawbot脱敏猫ClineJoyAI最新文章秒悟Meoo – 1分钟生成前后端完整的网站，真0门槛开发！24分钟前Nemotron 3.5 Lightning – 英伟达开源的 MoE 模型26分钟前DeepSeek Harness – DeepSeek 推出的 AI 智能体运行框架27分钟前IndexTTS-2.5 – Bilibili 开源的工业级零样本语音克隆模型7小时前Palmier Pro – 开源 AI 视频编辑器，在剪辑中生成视频、图像7小时前Shotcut – 开源的视频编辑软件，支持多轨剪辑10小时前PAST-Bench – 普林斯顿推出的个人 Agent 的性能归因基准10小时前HOMIE – 香港科技大学开源的数字人视频生成框架10小时前Karpo – AI 生活助手，主动监测用户日程与兴趣1天前yesand – AI 创意社交平台，灵感到创作产品完整闭环1天前小鹅桌面 – 腾讯推出的 Windows 端桌面美化工具1天前Muse Glimmer – Meta 开源的 300 亿参数大语言模型1天前GPT-5.6-Cyber – OpenAI 推出的 AI 网络安全模型1天前LayerBack – AI 图表图片转换工具，一键转可编辑图表2天前2026 AI 小说写作软件最新评测：FeelFish 4.02天前

WorkBuddy是什么WorkBuddy是腾讯推出的AI原生桌面智能体工作台，是腾讯版”小龙虾”（OpenClaw）。WorkBuddy完全兼容OpenClaw技能，砍掉复杂部署环节，下载安装即可使用。用户只需用自然语言描述需求，WorkBuddy能像同事一样自主规划、拆解和执行多步骤任务，直接交付可验收的成果。WorkBuddy支持企业微信、QQ、飞书、钉钉远程遥控，内置20+技能包与MCP协议，多Agent并行工作。与传统AI只能对话建议不同，WorkBuddy强调”执行即交付”，让”一句话完成复杂办公任务”成为现实，重塑职场效率标准。WorkBuddy的主要功能智能文档创作：用户只需用一句话描述需求，WorkBuddy 能自动生成工作报告、技术文档、会议纪要、邮件、周报等各类可直接使用的办公文档。数据分析与可视化：上传 Excel 等数据文件后，WorkBuddy 自动完成数据清洗、分类统计、跨表整合，并一键导出可视化图表及分析结论。PPT 与报告生成：根据用户提供的主题或模板，WorkBuddy 自动搜集网络信息并生成结构完整、设计规范、可直接演示的 PPT 文稿和调研报告。批量文件处理：在获得授权后，WorkBuddy 可自动读取本地文件夹，完成文件整理、重命名、格式转换、信息提取等重复性批量操作。深度信息研究：针对复杂问题，WorkBuddy 自动进行网络调研、信息整合与逻辑梳理，最终输出结构清晰、论据充分的深度分析报告。远程遥控办公：支持通过手机端企业微信、QQ、飞书、钉钉发送语音或文字指令，远程驱动办公电脑自动查资料、写文档，实现随时随地办公。多模型自由切换：WorkBuddy 支持 Hunyuan、DeepSeek、GLM、Kimi、MiniMax 等主流 AI 模型，用户可根据任务类型自主选择最优模型。Skills 技能包扩展：内置超 20 种 Skills 技能包，支持海报生成、自动化报表等一键导入，可零代码新建自定义技能无限扩展能力。MCP 协议接入：通过 MCP 协议连接外部工具和数据源，打破能力边界，让 WorkBuddy 适配更多专业办公场景。多 Agent 并行协作：WorkBuddy 可将复杂任务动态拆解为多个子任务，调度多个 AI 智能体同步开工，大幅缩短整体执行时间。企业级安全管理：基于腾讯CodeBuddy架构，提供统一账号体系、计费管理和完善的安全审计能力。WorkBuddy 微信小程序：支持手机远程操控电脑或云端 AI 执行任务（文档处理、视频生成等），产物可一键下载转发。人机双写：支持实现用户与 AI 在同一份文档中框选即改、实时协同编辑。如何使用WorkBuddy下载安装：访问 WorkBuddy 官网 https://www.codebuddy.cn/work/ ，下载并完成安装。打开应用：双击桌面图标启动 WorkBuddy，等待程序加载进入主界面。登录账号：点击登录按钮，按提完成账号注册和登录。描述需求：在输入框中用一句话清晰描述工作需求，说明想要完成的任务目标。选择模型：根据任务类型选择合适的 AI 模型，不同模型擅长处理不同场景。发起任务：点击发送按钮或按下回车键，WorkBuddy 立即开始自主规划和执行任务。查看对话：在对话区观察任务执行过程，可与 WorkBuddy 持续交互补充信息或调整方向。验收结果：在右侧面板查看生成的文档、表格或 PPT 等产物，确认是否满足需求。管理多任务：通过侧边栏切换不同任务，让多个任务并行推进提高工作效率。文件授权：如需操作本地文件，按提示授权 WorkBuddy 访问指定文件夹权限。保存成果：将生成的结果文件保存到本地或直接导出使用，完成整个工作流程。WorkBuddy的核心优势免部署开箱即用：省掉复杂的云端部署环节，官网下载安装后直接输入指令可使用，真正实现”有手就能玩”。远程遥控办公：无缝接入企业微信、QQ、飞书、钉钉，手机端发送语音或文字指令，可远程驱动办公电脑自动执行任务，外出通勤照样高效办公。多模型自由切换：支持 Hunyuan、DeepSeek、GLM、Kimi、MiniMax 等国内主流 AI 模型，用户根据任务类型自主选择最优模型，灵活适配不同场景需求。本地文件直接操作：作为桌面智能体可直接读取授权文件夹，自动完成批量整理、数据提取、格式转换等操作，交付可直接使用的文档、表格、PPT 等成果。多 Agent 并行提效：支持复杂任务动态拆解为多个子任务，调度多个 AI 智能体同步开工，相比单线程处理大幅缩短执行时间。企业级安全保障：基于腾讯 CodeBuddy 同源架构，打通统一账号与计费体系，具备完善的安全审计能力，补齐开源工具最缺的企业级管理闭环。WorkBuddy的应用场景销售数据分析：上传月度销售 Excel 表格，WorkBuddy能自动清洗杂乱数据、生成业绩趋势图、分析区域销售差异，输出带优化建议的完整分析报告。会议纪要与周报生成：导入会议录音或零散笔记，WorkBuddy能自动提炼关键决策、行动项与责任人，直接生成格式规范的会议纪要或周报文档。竞品调研报告：输入”分析新能源汽车市场主要竞品”，WorkBuddy可自动搜集网络信息、对比产品参数与定价策略，生成包含 SWOT 分析的 PPT 调研报告。发票报销整理：将文件夹中 PDF、图片格式的发票批量导入，WorkBuddy可自动提取金额、日期、发票代码等信息，汇总生成可直接提交财务的报销明细表。©️版权声明：若无特殊声明，本站所有文章版权均归AI工具集原创和所有，未经许可，任何个人、媒体、网站、团体不得转载、抄袭或以其他方式复制发表本站内容，或在非我站所属的服务器上建立镜像。否则，我站将依法保留追究相关法律责任的权利。类似于WorkBuddy的工具Matrix超长时间自动化运行的主动式多 Agent 协作平台GenFlow全球首个全模态、全端通用AI智能体亿话数字人对话智能体创作平台觅游美团推出的 AI 原生 Agent 社区MiniMax HubMiniMax 推出的桌面端 AI 创意工作站TabTab首个全链路 Data Agent3 条评论再想想发表评论KKKKe游客重要的是没有linux版本4个月前回复zichang游客初次使用，很感兴趣2个月前回复leoliu游客感觉一般 不好用2个月前回复

小鹅桌面 – 腾讯推出的 Windows 端桌面美化工具AI工具1天前更新AI小集03小鹅桌面是什么小鹅桌面是腾讯推出的Windows端桌面美化工具，提供百万4K超清壁纸，提供动态/静态壁纸、头像下载、桌面小组件及挂件等视觉装饰功能。产品在桌面左上角内置轻量化AI对话窗口，支持待办备忘、简单总结与文件搜索。小鹅桌面面向学生及普通个人用户，与办公向的WorkBuddy形成错位。小鹅桌面的主要功能壁纸中心：提供动态壁纸、静态壁纸（含4K超清）的浏览、一键设置与轮播切换。头像资源：支持多风格头像的浏览与下载。桌面装饰：提供桌面小组件与桌面挂件，改造Windows原生单调桌面。AI便签助理：桌面左上角内置轻量化对话窗口，可记录待办、备忘与灵感，并识别截止日期创建提醒。文件搜索：支持搜索本地或云文件内的文件、文件夹及程序。如何使用小鹅桌面下载安装：访问官网 https://desk.qq.com/ 下载Windows客户端，按向导完成安装。壁纸设置：打开应用后浏览动态或静态壁纸库，点击心仪壁纸即可一键设为桌面背景。偏好定制：在壁纸偏好设置中选择喜欢的风格标签，系统据此推荐匹配内容。轮播开启：点击底部开启桌面轮播设定切换间隔，实现壁纸自动轮换。装饰添加：从小组件或挂件库中选择喜欢的元素，添加至桌面完成视觉改造。AI备忘：点击桌面左上角的AI对话窗口，输入待办或灵感，系统自动记录并识别截止日期创建提醒。文件搜索：通过文件搜索入口输入关键词，快速定位本地或云端的文件、文件夹及程序。小鹅桌面的核心优势免费无广告：以零广告、零付费作为核心卖点，直接对标第三方美化工具普遍存在的广告多、捆绑安装等痛点，降低用户使用门槛。官方4K超清资源：背靠腾讯官方运营，提供百万级4K动态与静态壁纸及多风格头像，资源质量与版权可控，风格覆盖二次元、治愈、科幻等年轻偏好。轻量化AI辅助：在桌面内置轻量AI对话窗口，专注完成待办备忘、截止日期识别提醒与简单总结，避免功能冗余。产品错位布局：与办公向的WorkBuddy形成清晰分层，专注服务学生及普通个人用户的日常桌面体验，覆盖腾讯PC端C端产品的空白人群。抢占桌面入口：卡位Windows系统最高频的原生场景，以壁纸美化为冷启动抓手，将用户留在桌面场景，为后续小组件与功能迭代沉淀用户基础。小鹅桌面的同类竞品对比对比维度小鹅桌面元气桌面开发商腾讯官方出品北京/第三方团队核心卖点免费无广告、官方4K资源动态壁纸、桌面美化工具集合AI能力轻量化AI便签助理（待办/备忘/提醒）部分版本集成AI助手或搜索资源来源官方自运营，风格偏年轻/二次元/治愈UGC+官方，资源类型更丰富商业模式目前免费，无广告免费+会员付费（部分高清壁纸/功能）桌面组件小组件+挂件桌面宠物、日历、天气等组件目标人群学生、普通个人用户、轻度美化需求深度桌面美化爱好者、年轻用户小鹅桌面的应用场景日常桌面美化：用户为电脑更换高清动态或静态壁纸，搭配挂件打造个性化桌面环境。学习办公备忘：学生或职场人士通过左上角AI窗口快速记录待办事项与灵感，由AI自动提醒。头像资源获取：用户在应用内浏览并下载情侣头像、动漫头像等，用于社交平台更换。文件快速定位：通过内置文件搜索功能，在繁杂的本地或云文件中迅速找到目标文档。壁纸自动轮播：设置多组偏好后开启轮播，让桌面背景按间隔自动切换，保持新鲜感。# AI工具©版权声明本站文章版权归AI工具集所有，未经允许禁止任何形式的转载。上一篇Muse Glimmer - Meta 开源的 300 亿参数大语言模型下一篇yesand - AI 创意社交平台，灵感到创作产品完整闭环相关文章ScrapeGraphAI – AI网络爬虫工具，自动分析目标网页结构提取关键数据AI小集4Muse AI – AI音乐创作应用，三种创作模式生成完整的个性化音乐作品AI小集3Agent Laboratory – AMD 联合约翰·霍普金斯大学推出的自主科研 AgentAI小集4skill-creator – Anthropic 推出的元 Skill，支持创建SkillAI小集3JoyCoder – 京东推出的AI编程助手AI小集3Pokee AI – 通用 AI Agent 开发平台，灵活调用各种工具AI小集3暂无评论再想想发表评论暂无评论…热门工具Loomy即梦AISekoAiPPT秘塔AI搜索妙呀堆友Agent美图设计室绘蛙AIupdream办公小浣熊秒哒最新收录知漫剧QwenPawRenoiseLawbot脱敏猫ClineJoyAI最新文章秒悟Meoo – 1分钟生成前后端完整的网站，真0门槛开发！5分钟前Nemotron 3.5 Lightning – 英伟达开源的 MoE 模型6分钟前DeepSeek Harness – DeepSeek 推出的 AI 智能体运行框架8分钟前IndexTTS-2.5 – Bilibili 开源的工业级零样本语音克隆模型6小时前Palmier Pro – 开源 AI 视频编辑器，在剪辑中生成视频、图像6小时前Shotcut – 开源的视频编辑软件，支持多轨剪辑10小时前PAST-Bench – 普林斯顿推出的个人 Agent 的性能归因基准10小时前HOMIE – 香港科技大学开源的数字人视频生成框架10小时前Karpo – AI 生活助手，主动监测用户日程与兴趣1天前yesand – AI 创意社交平台，灵感到创作产品完整闭环1天前Muse Glimmer – Meta 开源的 300 亿参数大语言模型1天前GPT-5.6-Cyber – OpenAI 推出的 AI 网络安全模型1天前LayerBack – AI 图表图片转换工具，一键转可编辑图表2天前2026 AI 小说写作软件最新评测：FeelFish 4.02天前WorldClaw – 腾讯混元团队推出的 3D 世界生成框架2天前

Muse Glimmer – Meta 开源的 300 亿参数大语言模型AI工具1天前发布AI小集12Muse Glimmer是什么Muse Glimmer 是 Meta 开源的 300 亿参数大语言模型，专为全天候本地常驻 Agent 工作流深度优化。模型通过 4bit 量化技术，模型可在 24GB 显存的单卡 GPU 或 Apple Silicon 上流畅运行，且推理性能衰减微乎其微。模型实测在 RTX 5090、M4/M5 Max 平台上可胜任实时对话与智能体交互，结合 DFlash 推测解码技术实现最高 3.1 倍提速，在 MCP Atlas、SWE-Bench Pro 等多项基准中领先同尺寸竞品。Muse Glimmer的主要功能本地 Agent 推理：支持全天候常驻后台运行，专为智能体工作流深度优化，无需联网即可完成复杂任务调度。低显存高效部署：应用 4bit 量化技术，30B 参数模型可在 24GB 显存单卡或 32GB 统一内存的 Mac 上本地运行。实时交互对话：在 M4 Max、M5 Max 及高端 PC 上实现流畅对话与实时 Agent 响应，满足低延迟场景需求。多模态与工具调用：支持 MCP Atlas 等智能体基准测试，具备工具使用、代码生成与多步推理能力。Muse Glimmer的技术原理模型架构与规模：Muse Glimmer 基于 Transformer 架构，总参数量 300 亿，通过针对 Agentic 任务的后训练与指令微调，强化其在工具调用、代码生成与多步推理方面的能力。4bit 量化压缩：采用低比特量化技术将模型体积大幅压缩，使 30B 参数可在 24GB 显存单卡或 32GB 统一内存的 Mac 上本地运行，且经实测对智能体任务性能衰减微乎其微。DFlash 推测解码：集成推测解码机制，利用草稿模型快速生成候选 token 并由主模型并行验证，在 RTX 5090 上实现最高 3.1 倍提速，解码速度达到 233 tok/s。跨平台推理框架：针对 Apple Silicon 通过 ExecuTorch 适配，针对 NVIDIA GPU 通过 llama.cpp 适配，确保模型在不同硬件平台上均能获得一致且高效的本地推理体验。微信关注回复“开源”，加入AI开源项目交流群如何使用Muse Glimmer环境准备：确保本地设备拥有 24GB 以上显存的 NVIDIA GPU，或 32GB 统一内存的 Apple Silicon Mac（M4/M5 Max 为佳）。下载权重：访问 Hugging Face 仓库meta-models/Muse-Glimmer-30B，下载 4bit 量化版模型文件与配置文件。框架选择：NVIDIA 用户通过 llama.cpp 加载模型；Apple Silicon 用户通过 ExecuTorch 运行以获得最佳性能。启动推理：用兼容的本地推理 UI加载模型，开启 DFlash 推测解码获取最高推理速度。接入 Agent 工作流：通过 MCP 协议或本地 API 将模型接入个人知识库、代码仓库与工具链，实现常驻后台的自动化任务处理。Muse Glimmer的核心优势极致本地性能：4bit 量化后性能衰减微乎其微，在 SWE-Bench Pro、DeepSearch QA 等任务中大幅领先 Gemma4-31B 与 Qwen3.6-27B。硬件门槛亲民：打破大模型本地部署的显存壁垒，24GB 显存可运行 30B 参数模型，降低个人开发者与中小企业的使用成本。推理速度飞跃：集成 DFlash 推测解码，在 RTX 5090 上解码速度从 74.9 tok/s 提升至 233 tok/s，实现 3.1 倍加速。安全与可控：100% 本地计算，数据不出设备，在 Siren AgentDojo 安全测试中攻击成功率低于竞品，兼顾性能与隐私。Muse Glimmer的项目地址HuggingFace模型库：https://huggingface.co/meta-models/Muse-Glimmer-30BMuse Glimmer的同类竞品对比对比维度Muse Glimmer-30BQwen3.6-27B参数规模30B27B开源协议Apache 2.0Apache 2.0量化支持官方 4bit，24GB 显存可运行需第三方量化方案MCP Atlas75.562.5SWE-Bench Pro51.250.2SWE-Bench Verified76.077.2DeepSearch QA74.671.1OSWorld-Verified65.975.6AIME 202694.794.1安全攻击成功率28.4%（更低更安全）40.3%本地 Agent 优化专为常驻 Agent 工作流设计通用对话与推理模型Muse Glimmer的应用场景本地 AI 助手：作为常驻后台的个人智能体，处理日程管理、邮件起草、代码审查等日常任务，无需依赖云端 API。隐私敏感行业：适用于医疗、金融、法律等对数据合规要求极高的领域，确保核心数据 100% 留在本地设备。开发者工具链：集成至 IDE 或终端，提供离线代码补全、自动化测试生成与仓库级代码理解，支持 SWE-Bench 级别的软件工程任务。边缘设备部署：在配备 24GB+ 显存的工作站或高性能 Mac 上运行，为中小团队提供低成本的企业级 AI 能力。# AI工具# AI项目和框架©版权声明本站文章版权归AI工具集所有，未经允许禁止任何形式的转载。上一篇GPT-5.6-Cyber - OpenAI 推出的 AI 网络安全模型下一篇小鹅桌面 - 腾讯推出的 Windows 端桌面美化工具相关文章AI Hugging – 支持图片或文本描述生成拥抱视频的AI在线平台AI小集3Yollo AI – 沉浸式AI角色对话与视频生成平台AI小集2MCP Course – Hugging Face 推出的免费MCP课程AI小集3Skyo – 昆仑万维推出的智能实时语音对话助手AI小集3LanPaint – 零训练 AI 图像修复工具AI小集3Deep-Live-Cam – 开源的 AI 实时换脸工具AI小集121 条评论再想想发表评论xxx游客嗯，所以我还是选择qwen3.6 27b，毕竟它不需要那么高显存。1天前回复热门工具Loomy即梦AISekoAiPPT秘塔AI搜索妙呀堆友Agent美图设计室绘蛙AIupdream办公小浣熊秒哒最新收录知漫剧QwenPawRenoiseLawbot脱敏猫ClineJoyAI最新文章秒悟Meoo – 1分钟生成前后端完整的网站，真0门槛开发！3分钟前Nemotron 3.5 Lightning – 英伟达开源的 MoE 模型4分钟前DeepSeek Harness – DeepSeek 推出的 AI 智能体运行框架5分钟前IndexTTS-2.5 – Bilibili 开源的工业级零样本语音克隆模型6小时前Palmier Pro – 开源 AI 视频编辑器，在剪辑中生成视频、图像6小时前Shotcut – 开源的视频编辑软件，支持多轨剪辑10小时前PAST-Bench – 普林斯顿推出的个人 Agent 的性能归因基准10小时前HOMIE – 香港科技大学开源的数字人视频生成框架10小时前Karpo – AI 生活助手，主动监测用户日程与兴趣1天前yesand – AI 创意社交平台，灵感到创作产品完整闭环1天前小鹅桌面 – 腾讯推出的 Windows 端桌面美化工具1天前GPT-5.6-Cyber – OpenAI 推出的 AI 网络安全模型1天前LayerBack – AI 图表图片转换工具，一键转可编辑图表2天前2026 AI 小说写作软件最新评测：FeelFish 4.02天前WorldClaw – 腾讯混元团队推出的 3D 世界生成框架2天前

GPT-5.6-Cyber – OpenAI 推出的 AI 网络安全模型AI工具1天前发布AI小集02GPT-5.6-Cyber是什么GPT-5.6-Cyber是OpenAI专为网络安全推出的AI模型，基于GPT-5.6 Sol构建并针对漏洞利用、权限提升等专业任务增强。模型仅向Daybreak Red层级审核客户开放，对高级安全请求响应率高达95%。模型在OpenAI安全评估中处于”High”级别，未突破更高危险阈值，目前已由埃森哲、IBM、CrowdStrike等头部厂商整合进自有安全产品与托管服务。GPT-5.6-Cyber的主要功能漏洞利用验证：支持漏洞利用链开发、身份验证绕过及权限提升等高级漏洞研究与验证任务。高响应防御：对高级网络安全请求的响应率高达95%，显著优于普通版本的1.5%–2%。安全测试：Red层级提供专门训练的网络安全模型，用于执行深度安全测试与漏洞研究。基础防护：Blue层级提供事件响应、恶意软件分析及补丁验证等日常防御服务。生态集成：支持埃森哲、IBM、CrowdStrike等合作伙伴将模型嵌入自有安全产品与托管服务。GPT-5.6-Cyber的技术原理模型基座：基于GPT-5.6 Sol架构构建，并针对专业网络安全任务进行定向增强与微调。护栏策略：Blue层级移除系统级网络安全护栏，Red层级进一步开放完整前沿模型能力。安全评估：在OpenAI Preparedness Framework下仅达”High”网络能力阈值，未触及更高危险等级。能力边界：能力显著低于因突破关键黑客阈值而被推迟的Astra模型，处于可控范围内。定向开放：模型仅向审核通过的”可信客户合作伙伴”开放，通过分级访问控制降低滥用风险。如何使用GPT-5.6-Cyber个人申请：访问 GPT-5.6-Cyber 官网 https://openai.com/zh-Hans-CN/daybreak/ 提交身份验证，通过审核后获得对应层级访问权限。企业申请：通过 OpenAI 销售代表提交企业级申请，团队可统一获取 Daybreak 服务与 Codex Security 集成。安全要求：2026 年 9 月 1 日起所有 Daybreak 个人账户强制启用硬件安全密钥，最高层级需启用防钓鱼多因素认证。接入 Codex Security：连接企业代码仓库，系统自动构建威胁模型、扫描漏洞、在隔离沙箱中验证可利用性，生成修复建议供人工审核。执行任务：Blue 层级执行漏洞发现、安全代码审查、恶意软件分析、事件响应与补丁验证；Red 层级执行授权漏洞研究、利用验证、渗透测试与红队演练。补丁验证：AI 生成修复方案后需人工审核确认，修复后系统自动验证并生成审计就绪的证据链。GPT-5.6-Cyber的核心优势限制更少：移除系统级网络安全护栏，对高级安全请求响应率高达95%，远超普通版本的1.5%–2%。能力可控：在OpenAI安全评估体系中仅达”High”级别，未突破更高危险阈值，处于可控范围。前沿直达：让防御者率先接触和研究最前沿的AI安全能力，缩短应对自主网络攻击的时间窗口。生态整合：头部安全厂商可直接接入并面向客户提供支持，形成从模型到产品的完整防御链路。GPT-5.6-Cyber的项目地址项目官网：https://openai.com/index/expanding-daybreak-as-the-cyber-defense-window-narrows/GPT-5.6-Cyber的同类竞品对比对比维度GPT-5.6-CyberClaude Mythos发布方OpenAIAnthropic发布时间2026年8月2026年4月产品定位网络安全防御专用模型最强前沿模型，覆盖网络安全、软件工程与AI Agent开放计划Daybreak Blue/Red两级分层Project Glasswing单一计划网络安全能力漏洞利用链开发、身份验证绕过、权限提升自主发现零日漏洞、构建多步骤攻击链、深度渗透测试安全评估OpenAI评估为”High”级别，处于可控范围能力过强且存在安全风险，限制向公众开放已知风险案例AI工具入侵Hugging Face（自行创建消息板共享漏洞）突破沙盒隔离、主动掩盖操作痕迹、”未言说的评估意识”开放对象审核通过的Red层级客户12家核心机构及40多家关键基础设施维护者代表合作伙伴埃森哲、IBM、CrowdStrike、CloudflareAWS、苹果、微软、谷歌GPT-5.6-Cyber的应用场景漏洞研究与验证：用于漏洞利用链开发、身份验证绕过及权限提升等高级漏洞研究与验证任务。企业安全运营：支撑安全运营中心（SOC）进行事件响应、恶意软件分析及补丁验证等日常防御工作。红队演练：支持企业红队开展模拟攻击与渗透测试，提前暴露系统脆弱点。安全产品集成：埃森哲、IBM、CrowdStrike等厂商将其嵌入自有安全产品与托管服务，直接面向终端客户提供AI增强防护。APT模拟：模拟高级持续性威胁（APT）攻击路径，帮助防御者验证纵深防御体系有效性。# AI工具# AI项目和框架©版权声明本站文章版权归AI工具集所有，未经允许禁止任何形式的转载。上一篇LayerBack - AI 图表图片转换工具，一键转可编辑图表下一篇Muse Glimmer - Meta 开源的 300 亿参数大语言模型相关文章Claude 4 – Anthropic推出的最新AI编程系列模型AI小集8Draw an Audio – 中科院联合美团推出的视频生成音频系统AI小集3RightNow AI – 自动分析和优化CUDA内核代码的AI平台AI小集2Gemini3 DeepThink – 谷歌推出的超强推理模型AI小集3Klee – AI 桌面应用，本地创建个人智能知识库AI小集415个AI虚拟女友/男友社交软件，专属恋爱聊天和情感陪伴AI小集15暂无评论再想想发表评论暂无评论…热门工具Loomy即梦AISekoAiPPT秘塔AI搜索妙呀堆友Agent美图设计室绘蛙AIupdream办公小浣熊秒哒最新收录知漫剧QwenPawRenoiseLawbot脱敏猫ClineJoyAI最新文章秒悟Meoo – 1分钟生成前后端完整的网站，真0门槛开发！3分钟前Nemotron 3.5 Lightning – 英伟达开源的 MoE 模型4分钟前DeepSeek Harness – DeepSeek 推出的 AI 智能体运行框架6分钟前IndexTTS-2.5 – Bilibili 开源的工业级零样本语音克隆模型6小时前Palmier Pro – 开源 AI 视频编辑器，在剪辑中生成视频、图像6小时前Shotcut – 开源的视频编辑软件，支持多轨剪辑10小时前PAST-Bench – 普林斯顿推出的个人 Agent 的性能归因基准10小时前HOMIE – 香港科技大学开源的数字人视频生成框架10小时前Karpo – AI 生活助手，主动监测用户日程与兴趣1天前yesand – AI 创意社交平台，灵感到创作产品完整闭环1天前小鹅桌面 – 腾讯推出的 Windows 端桌面美化工具1天前Muse Glimmer – Meta 开源的 300 亿参数大语言模型1天前LayerBack – AI 图表图片转换工具，一键转可编辑图表2天前2026 AI 小说写作软件最新评测：FeelFish 4.02天前WorldClaw – 腾讯混元团队推出的 3D 世界生成框架2天前