与「TencentAI」相关的搜索结果

最近当大家都在刷屏Fable 5和GPT-5.6 的时候， 殊不知腾讯已经悄无声息的把大模型能力给追上来了。 你们知道腾讯低调到什么程度吗？ 前几天发布的Hy3 ，21B的激活参数已经可以打平旗舰水准，并且直接塞进微信10 亿+用户手里 ， 这才是国产大模型真正的王炸啊兄弟们！ 说实话周一刷到 Hy3 发布消息那会儿， 我第一反应跟很多人一样：又一个卷参数的国产模型？ 趁着周末有空，我仔细翻了一圈资料， 发现完全不是这么回事。 原来@TencentHunyuan这个模型从一开始就不是朝着聊天模型去的，人家的定位就是奔着干活去的 Agent 向 LLM。 另外最反直觉的是它的推理效率，295B 总参数， 实际激活只有 21B 的 MoE 架构， 就能打平参数是它 2 到 5 倍的旗舰模型， 这事儿我琢磨了一下，其实有点像团队管理， 并不是说人越多产出越高， 关键得看真正干活的那几个人靠不靠谱。 为什么这个效率重要呢？ 你想想微信每天服务 10 亿人， token 成本多一厘都是天文数字， 21B 激活还能打出旗舰水准， 我觉得这才是它能被规模化铺开的根本原因， 跟刷榜没半毛钱关系。 说到刷榜，我仔细研究之后这个版本还真不是， 从 preview 到正式版，Hy3 是Hunyuan团队拿了50 多个真实业务的反馈不断迭代出来的，内部 WorkBuddy 任务成功率从 72% 干到 90%，耗时降了 34%，幻觉和常识错误全往下掉。 我原来以为 Agent 的核心是想得多深， 后来发现真正决定能不能上生产的，是稳不稳， 比如工具编排会不会中途崩，出错了能不能自己爬回来，换个框架差不差太多等等， 就是这些不起眼的东西， 才是能跑完一整套 workflow 不掉链子的关键。 当然实际测试下来Hy3 也有短板， 比如纯视觉它还不够强，不过腾讯也有单独的多模态大模型线，它能打的主要还是 coding、办公、复杂任务规划这几块。 最后说个我的一个判断：就是接下来大模型拼的可能不是谁参数更大了，核心看谁能把干活的能力以最低摩擦送到更多人手里，比如Hy3 我理解它可能不是一个更会聊天的模型，更像是一个更会干活的 Agent 底座。 话说回来，腾讯这步棋确实走得挺狠， 我觉得国产大模型的拐点真的要到了， 下方视频是我用Hy3跑的几个测试案例， 分别让它做了一个介绍自己的HTML网页， 一个我的认知系统agent网页， 和做了一个10页的PPT，都很惊艳，大家可以感受下， 会干活、会自检、还会主动说自己哪没做到位， 我觉得这才是 Agent 时代，真正能用的模型该有的样子hh～ #Hy3# #Hunyuan# #TencentAI#

人类止步！腾讯 @TencentAI_News 推出仅供 AI 智能体 🤖 使用的 #AgentlyMail，可以让# AI 智能体收发邮件。 邮箱只允许 AI 智能体使用，人类可以创建随机邮箱、管理账号，但不能手动写邮件，前台需要通过微信扫码绑定，每个账户最多创建两个随机邮箱，不可指定邮箱前缀：

🔥 7 天狂揽 2T 吞吐！ 免费模型活动持续进行中！ 算力狂欢持续升温！活动上线仅 7 天， 全平台累计 Token 吞吐量已正式跨越 2 万亿（2T）历史级里程碑！面对高并发洪峰与海量自动化调度， 算力底座稳如磐石！ 4 大国产热门模型全量免费，零门槛开箱即用： 🔸MiMo-V2.5：最新加入，解锁全新推理体验 🔸DeepSeek-V4-Flash：极速响应，高并发任务首选 🔸DeepSeek-V4-Flash-Vision-Exp：全功能多模态视觉理解 🔸Tencent Hy3：扎实中文能力与通用推理底座 不管你是重度 Coding、跑超长上下文，还是搭建企业级高并发 Agent 智能体——算力管够， 替你买单！ 👉 立即开冲：

XPENG’s robotics business has completed a 900M+ Series A funding round at a post-money valuation exceeding $6.3B, setting a new fundraising record for China’s embodied AI industry. The round was led by IDG Capital, joined by Gaorong Ventures, and strategically backed by Tencent and Alibaba, with XPENG retaining controlling ownership. The capital will accelerate full-stack hardware/software R&D, physical AI training, smart manufacturing infrastructure, and global market expansion. Backed by 12 years of in-house tech innovation—from physical foundation models to Turing AI chips and robust AI infrastructure—XPENG is uniquely positioned to lead the scale-up and commercial rollout of high-end humanoid robots. XPENG IRON is slated for mass production in late 2026, launching first in XPENG retail stores and campuses before rolling out globally with customer deliveries in 2027. $XPEV

免费模型继续加码, 冲刺单日万亿吞吐量 Q3 已经达成 1T Token 吞吐量 但看起来还没准备停下 今天 MiMo-V2.5 也加入免费阵营 再加上这几天陆续开放的 · DeepSeek-V4-Flash · DeepSeek-V4-Flash-Vision-Exp · Tencent Hy3 免费模型越堆越多 Web Chat 和 API 双端免费 真实调用量也只会继续往上走 看这个节奏 应该是准备迈入万亿俱乐部了 @justinsuntron @BAI_AGI #TRONEcoStar#

Should enterprise AI follow a workflow — or think for itself? The answer is: both. Real business work is a mix of open-ended problems and predictable processes. That’s why Claw Mode in Tencent Cloud ADP 4.0 lets Agents and Workflows call each other. 🔄

a life update: i started a new company called Pragmatik (p7k) Labs (语用科技) in shanghai, focusing on the research of next-generation agents across digital and physical worlds. thanks to Gaorong Ventures and HSG (红杉中国 & 高榕创投) for co-leading this round, and to Tencent (腾讯) and Shanghai Engine Fund (上海未来产业基金) for the support. @pragmatik_labs ·

A bigger WorkBuddy workspace starts today. Our refreshed Personal Plans are now live, with more credits and more flexibility for the work you already bring to WorkBuddy. And through August 31, 2026 (PT), Hy3 from @TencentHunyuan is free to use in WorkBuddy Global. ✨ Here’s what you get: More room to work. Pro monthly credits have doubled from 1,000 to 2,000, giving you more capacity for multi-step tasks, longer workflows, and the inevitable “one more revision.” More credits for staying active. For a limited time, free users can earn 30 extra credits per active day, while Pro users can earn 50. With daily rewards, that’s up to 1,000 credits/month on Free and up to 3,500/month on Pro. More control over how you work. Pro gives you access to all available models, so you can choose based on the task—whether you’re looking for a quick response or working through something more complex. More room to code. Unlimited code completion comes with Pro, helping you stay in flow across longer coding sessions. More flexibility on price. Pro is $10/month, or $96/year—equivalent to $8/month with annual billing. More ways to get started. New users receive 250 welcome credits at sign-up, and eligible users can unlock a one-time 7-day Pro trial with 500 trial credits. Plus, Hy3 is free through the end of August—so this is a good time to bring over the work that’s been waiting: research, reports, decks, code, analysis, and everything in between. See the new plans ↓

腾讯做的这个团队记忆 Tencent Agent Memory 可以让团队里的多个 AI Agent 共享记忆 那多个 Agent 共用一份团队记忆后，如果写进去的是错误信息怎么办？

Minimax H3 is the first *open* video model I've seen that outperforms HunyuanVideo 1.5 - which is actually an impressive accomplishment for @TencentHunyuan to have held that throne for so long. For a while now I've been sad that the image/video frontier seemed to be getting much more closed than the text LLM frontier. H3 is a very strong step back in the open direction. This was generated in ~30 min on my AMD laptop: