Discover the AI Tools

智谱今日发布GLM-5.3，与上代不同，这次基础模型保持不变，所有提升都来自后训练阶段的扩展。通过十倍规模的长程任务环境、更多样的环境类型和更长的训练时间，模型智能上限被显著拉高。内部评测显示，编程体验比GLM-5.2提升约50%，成为当前编程能力最强的开源模型。在公开基准测试中，GLM-5.3的编程和智能体能力已接近Claude Fable 5，并在多个开源榜单中登顶。

Apple is teaming up with Alibaba to develop a dedicated large language model for the Chinese market, marking a strategic shift from its previous approach. This collaboration aims to bring Apple Intelligence to China while navigating regulatory hurdles. The move positions Apple as a pioneer in dual-track AI deployment, blending local expertise with global innovation.

腾讯大模型团队迎来关键人事调整，资深研究员许灿从腾讯元宝转至微信事业群，加入WeLM团队，专注后训练与Agent开发。许灿曾创建WizardLM项目，提出Evol-Instruct方法，在合成数据、强化学习反馈等领域经验丰富。此次调整正值微信AI加速落地，其AI助手“小微”已开始灰度测试，WeLM团队也在推进稀疏MoE架构以降低推理成本。许灿的加入有望增强微信AI的任务规划与工具调用能力，为腾讯在C端AI入口竞争中开辟新路径。

The U.S. Department of Justice is pushing to ban Google from paying browsers for default search status, a move that could devastate Mozilla, which relies on these payments for 85% of its U.S. revenue. This ironic twist in the antitrust battle may weaken competition rather than strengthen it, leaving Firefox's future hanging in the balance.

贾樟柯的新片《敦煌妈妈》正式立项，讲述一位独居母亲在敦煌与AI相伴，最终踏上横跨中国的旅程。影片由赵涛、廖凡主演，预计2027年上映。这一设定将AI带入艺术电影的核心叙事，探讨科技与孤独、连接与疏离的复杂关系。贾樟柯以一贯的写实风格，为AI焦虑的时代提供了一抹温暖的人文视角。

智谱AI于8月14日发布GLM-5.3大模型，参数规模仍为740亿，但通过后训练优化，性能较上代提升50%，在多项基准测试中创下开源模型新高。其编程与智能体能力接近Claude Fable 5，编程体验超越国内其他模型。GLM-5.3已在官方工具及第三方平台开放，API即将上线，完整权重将在两周内开源。

At Baidu's AI Day, the company unveiled the Chinese name for its AI agent GenFlow: Kuku AI. With over 100 million monthly active users, Kuku AI is now available as a PC client, web version, mini program, and enterprise edition, signaling Baidu's push to dominate the AI office space.

Kingsoft Office's AI assistant, Lingxi Professional Edition, has officially integrated DeepSeek's latest flagship model, V4-Pro, bringing support for up to 1 million tokens of context. This means the tool can handle complex office tasks with unprecedented depth, understanding your documents, meetings, and preferences. The update is available now, and users can switch to the new model manually. It's a significant step toward more autonomous and reliable AI-powered office work.

京东七鲜咖啡宣布全球首家24小时智能无人咖啡店将于8月16日在北京银河Soho开业。店内从制作到配送全由机器人和AI完成，顾客还能自由调整咖啡浓度和甜度，甚至自定义饮品名称。开业当天将进行24小时机器人制作直播，并挑战吉尼斯世界纪录。更值得期待的是，AI个性化推荐功能即将上线，能根据用户身体状况建议咖啡因摄入量。这一模式能否持续吸引顾客，仍需市场检验。

本地部署的模型未必能轻松应对真实业务。为此，蚂蚁数科团队与开源社区合作，维护了本地大模型后训练工具AReno，并与百灵大模型联手，在单机环境下成功构建了Agentic RL循环。通过井字棋验证，模型在工具调用和动作选择上实现了质的飞跃，为实际应用铺平了道路。

DeepSeek's latest model, V4 Pro, has landed on SiliconFlow with a million-token context window and three reasoning intensity levels. The standout feature? Cache-hit tokens cost just $0.44 per million, a game-changer for agent workflows that rely on repeated calls. This launch also emphasizes coding and tool use, keeping the MIT license intact.

苹果 HomePod mini 发布近六年，终于要迎来更新。第二代产品预计今秋亮相，搭载更快的芯片以支持新一代 Siri 的本地 AI 计算。同时，UWB 和音质也将升级。这款入门级智能音箱能否扭转 Siri 的弱势形象，值得期待。

8月14日，一汽-大众全新纯电车型ID. AURA T6开启盲订，并确认接入豆包大模型。这标志着双方在智能座舱领域的合作正式扩展到纯电领域。此前，该合作已覆盖品牌全部燃油车型。

Launched on August 10, the Qwen Open Platform has already drawn over 200 ecosystem partners, including big names like SF Express and KE Holdings. The platform lets developers integrate AI agents into the Qwen app, covering everything from logistics to local life. With built-in tools and low barriers, it's quickly becoming a hub for AI-driven services. But what does this mean for the future of AI agents? Let's dive in.

MiniMax has unveiled Music3, a new AI model that can generate complete songs up to five minutes long from just lyrics and a description. Using a clever two-part architecture, it handles everything from the overall structure to the finest audio details. The result is a high-quality stereo track that could change how music is made.

微软宣布，Copilot语音模式中那个会做表情的黄色小怪物Mico将不再出现在聊天界面，而是被转移到Learn Live学习平台。这个决定背后是Copilot与Microsoft 365 Copilot应用的合并计划。Mico从推出到退役不到一年，虽然短暂，但它为AI语音交互的“人性化表达”提供了宝贵参考。微软的虚拟助手们——从回形针Clippy到语音助手Cortana，再到现在的Mico——似乎都难逃被淘汰的命运。

8月14日，广东在广州海珠区推出首个面向代币经济的金融产品“Token贷”，为算力和AI相关的中小企业提供融资新渠道。以中国银行广州分行为例，其“算力Token贷”根据企业合同金额和Token消耗量确定贷款额度，支持信用、应收账款质押和订单融资等多种担保方式。目前，该行已试水审批，累计资金规模达2800万元。这一创新举措将算力、Token使用和订单数据纳入信用评估，为AI企业融资开辟了新维度。

Google has unveiled Gemini 3.7 Flash, an upgraded AI model that promises to transform coding and app development. With superior code debugging and production-ready output, it cuts token costs and boosts efficiency. Starting today, it powers Gemini Spark, making developers' lives easier. Here's what you need to know.

OpenAI has finally rolled out the 'Ultra-Fast' mode for GPT-5.6 Sol, promising a 14-fold speed boost. But this isn't just about raw speed—it's a clever hardware play with Cerebras. The catch? Access is limited, and only certain customers will get it. Here's what you need to know.

A developer has exposed a troubling trend on the App Store: apps using AI-generated fake screenshots to display inflated ratings, like a 4.9-star score, are slipping through Apple's review process. This raises serious questions about the effectiveness of Apple's quality checks and the potential privacy risks, especially for Safari extensions that can access user data.