Channel
@OneMillion_AI
On this record: Topic · Growth · Engagement · Reactions · Posts · Citations · Cite this entry
3,856subscribers
+645 since we began measuring on 6 August 2026
Risers and fallers across the register · movement among entries of 3,162–10,000.
| Telegram ID | -1003894768046 |
|---|---|
| Type | Channel |
| Username | @OneMillion_AI |
| Created | Between 1 February 2026 and 31 July 2026 — estimated from Telegram’s id allocation, not measured. How this range is calculated. |
| First recorded | 6 August 2026 |
| Last confirmed live | 27 September 2026 |
| Measurements held | 15 |
| Confirmed unchanged | 1 time, most recently 27 September 2026 |
| On Telegram | t.me/OneMillion_AI |
Technology — a classification, not a measurement. An on-box language model (Qwen3.6-35B-A3B-FP8, prompt version 1) read this channel’s own recent posts on 14 September 2026 and assigned it the closest of 31 fixed categories, at 99% confidence. This is a model’s judgement about what the channel is likely to be about, not a fact this register measured the way a subscriber count or a view count is measured — it can be revised on a later pass, and it carries no weight anywhere else on this page. How this classification works, and why it has no browse page of its own yet.
| Measured (UTC) | Subscribers | Change |
|---|---|---|
| 27 Sept 2026, 02:56 | 3,856 | +227 |
| 15 Sept 2026, 05:18 | 3,629 | +41 |
| 11 Sept 2026, 18:39 | 3,588 | +22 |
| 7 Sept 2026, 12:20 | 3,566 | +42 |
| 2 Sept 2026, 11:53 | 3,524 | +24 |
| 30 Aug 2026, 17:26 | 3,500 | +4 |
| 28 Aug 2026, 02:52 | 3,496 | +33 |
| 25 Aug 2026, 04:45 | 3,463 | +16 |
| 21 Aug 2026, 22:18 | 3,447 | +34 |
| 19 Aug 2026, 00:34 | 3,413 | +49 |
| 15 Aug 2026, 17:57 | 3,364 | +66 |
| 12 Aug 2026, 15:33 | 3,298 | +47 |
| 9 Aug 2026, 21:31 | 3,251 | +30 |
| 7 Aug 2026, 00:41 | 3,221 | +10 |
| 6 Aug 2026, 04:49 | 3,211 | first reading |
39 posts held, back to 5 August 2026 — the reader has not yet reached the start of this channel’s public history, so older posts may sit further back, unread. Read across 3 pages of Telegram’s post history, 20 posts per page.
Nothing published in the last 30 days. ERR and ER are rolling 30-day measures, so there is nothing to compute — we hold 39 posts for this entry, the most recent from 7 August 2026. An engagement rate over an empty window would be a number about nothing.
113 reactions across 19 posts, in 11 distinct kinds. The most used accounts for 48.7% of them.
| Reaction | Count | Share | Share, drawn |
|---|---|---|---|
| 🤡 | 55 | 48.7% | |
| 💩 | 16 | 14.2% | |
| 😁 | 13 | 11.5% | |
| 👍 | 11 | 9.73% | |
| ❤ | 7 | 6.19% | |
| 🤬 | 4 | 3.54% | |
| 👎 | 2 | 1.77% | |
| 🤩 | 2 | 1.77% | |
| 😨 | 1 | 0.885% | |
| 🤔 | 1 | 0.885% | |
| 🤷♂ | 1 | 0.885% |
No sentiment is inferred, and none should be read in. This table is ordered by count and by nothing else. Emoji do not carry stable meaning across languages or communities — 🙏 is thanks in one channel and mourning in another — so we publish which ones were pressed and how often, and pass no judgement on what an audience meant by them.
Precision. Telegram publishes reaction counts per emoji and short-forms each one — 4.34K, 1.2M — so any single kind at or above 1,000 reaches us at three significant figures, and only counts below 1,000 are exact. The shares above are ratios of those figures and carry the same error. This is also why the total here can differ slightly from a reaction total printed elsewhere on the page: both are sums of the same rounded parts, taken over samples with different edges.
Coverage. Reactions were read on 35 of the 39 sampled posts in this sample. Summed by Telegram’s own count on each post — not by adding up the per-emoji breakdown above — those same posts carry 163 reactions in total: the kind of figure the paragraph above means by “a reaction total printed elsewhere on the page”.
Measured over the 39 most recent posts we hold, published 5 August 2026 to 7 August 2026, using the newest reading held for each. Telegram Stars are excluded: they are a payment, not a reaction, and they have their own section.
Gemini 3.5 Pro一拖再拖，3.7 Flash已经开始冒头 Gemini 3.5 Pro 还没发出来，下一款 Flash 已经开始冒头。有爆料者称，谷歌正在为 Gemini 3.7 Flash 做准备，目前还没有发布时间。 另一则爆料称，Gemini 3.5 Pro 原计划 8 月 10 日发布，现在可能连 8 月也赶不上。谷歌最初计划 6 月推出这款旗舰模型，此后一直延期，目前仍在与合作伙伴测试。 此前彭博披露，3.5 Pro 主要卡在编程能力。谷歌 6 月底还更新了训练数据继续优化，但结果依然没有达到预期。 Dan
🤡11
SemiAnalysis唱衰谷歌DeepMind：已经不算最前沿AI实验室了 SemiAnalysis 最新报告直接判定，谷歌 DeepMind 已经不再是最前沿 AI 实验室，未来重新做到 SOTA 的概率也降到了零。 它认为，DeepMind 正同时失去顶尖人才和算力。Jeff Dean、Sanjay Ghemawat、Quoc Le 和 Oriol Vinyals 最新集体离开谷歌创业。此前 Gemini 联合负责人 Noam Shazeer 已转投 OpenAI，诺奖得主 John Jumper 加入 Anthropic。 算力也被大量卖给竞争对手。SemiAnalysis 估算，2026 年第三季度到 2027 年第四季度，超过 20% 的 TPU 出货会直接卖给 Anthropic，相当于把大量稀缺算力长期锁给 Gemini 的对手。 SemiAnalysis 最终把问题归到谷歌「官僚、缓慢、战略保守」的组织…
😁5👎1
蚂蚁发布Ling-3.0小模型，两项Agent评测超Qwen3.5-9B 蚂蚁百灵（Inclusion AI）发布 Ling-3.0-tiny。这是一款面向本地部署和 Agent 场景的小型 MoE 模型，共有 79 亿参数，每个 Token 只激活 13 亿参数；同时支持思考和快速回答两种模式，还能调用工具执行任务。 虽然每次只激活 13 亿参数，它在部分 Agent 任务上已经能超过更大的 Qwen3.5-9B。官方测试中，Ling-3.0-tiny 在 GDPval v2-AA 和 TAU3-Banking-AA 两项通用 Agent 评测上都拿到更高分。 Ling-3.0-tiny 已上线 OpenRouter 和 Vercel AI Gateway，目前可以免费试用，模型权重也将在之后开源。 蚂蚁百灵
💩5
一个人4个月做到2.5万Star，Agent版tmux进YC Herdr 加入 Y Combinator F26。它可以理解成专门给 Coding Agent 用的 tmux：把 Claude Code、Codex、Cursor 等开在里面，统一挂后台、切换和查看状态。终端关掉、SSH 断开，Agent 也能继续跑。 普通 tmux 只能管终端，Herdr 还知道哪个 Agent 正在干活、哪个等你回复、哪个已经完成。Agent 之间还能互相发消息、等对方跑完。随着开发者同时开越来越多 Coding Agent，这类工具就是专门解决「一堆 Agent 怎么管」的问题。 Herdr 此前一直只有 Can Celik 一个人开发。上线 4 个月，已经拿到约 2.5 万 GitHub Star、34 万次下载，插件市场一个月出现 500 多个插件。现在项目加入 YC，核心运行时继续免费开源，许可证也从 AGPL 放宽到 Apac…
👍2
AI下一战不只卷大模型，ARC-AGI创作者看好整套Agent系统 知名 AI 评测 ARC-AGI 的创作者 François Chollet 表示，今天的高性能 AI Agent，本质上已经是「模型 + 程序」组成的完整系统。大模型负责理解和判断，外层程序负责控制流程、调用工具和运行代码。 Chollet 把这种神经符号架构形容成「符号三明治」：程序驱动神经网络，神经网络再调用程序。他认为，大模型还会继续变强，但外层程序同样会越来越重要，复杂任务最终要靠两者配合完成。 更进一步，Chollet 预测，现在主要靠工程师手写的外层程序，未来也可能由 AI 自己学习和进化。未来 AI 比拼的可能不再只是模型，而是整套 Agent 系统。 Chollet
👍2
哪个客户端用DeepSeek最省钱？第一竟是智谱ZCode OpenCode 联合创始人 Dax Raad 晒出过去 48 小时 DeepSeek 流量的客户端缓存命中率。结果第一名不是自家 OpenCode，而是智谱的 ZCode，命中率高达 98.60%。OpenCode V2 为 97.86%，Claude Code / CLI 则只有 89.31%。 Dax 自己看到都懵了，直言「不知道 ZCode 是什么，但它干得真不错」。DeepSeek 的缓存未命中输入比命中贵约 50 倍，按同样的 Token 结构粗算，ZCode 这组命中率对应的输入成本只有 Claude Code 的约 27%。 ZCode 是智谱面向 GLM-5.2 推出的 Agentic Development Environment (ADE)，但也支持接入其他模型。 dax
👍3
MiniMax Code从桌面走进终端，CLI版本即将上线 MiniMax Code 研发负责人何涛预告，MiniMax Code 即将推出终端版。从晒出的测试界面看，它已经是完整的 Coding Agent 形态，可以在终端里理解任务、思考并执行操作。 MiniMax Code 本身并不是新产品。MiniMax 此前已经有 Web 和桌面版，今年 5 月底正式把桌面 App 改名为 MiniMax Code；6 月发布 M3 时，又把它定位为专门配合自家模型使用的 Agent 产品。官方称其底层 harness 基于 OpenCode 和 Pi。 何涛
💩6🤷♂1
前AV女优真「转码」了：退圈后用Claude做网站、搞自动化 日本前 AV 女优 Noa 在 X 上自曝，退圈后转行做了程序员。她现在用 Claude 做自动化、开发网站，还表示 Web 制作和自动化相关咨询可以直接私信她。 她称原来约 2 万粉丝的账号被事务所收走，只能重新起号。做 AV 女优时，她曾用过「菜月ひかる」「花吹詩音」等艺名。 消息传到程序员圈后，网友很快给这次转行起了个名字：「AV 转码」。 原帖
🤡8🤩2🤔1
Grok Build终于转正：1.0没憋大招，先把Beta标签撕了 Grok Build 发布 1.0 正式版，结束此前的 Beta 阶段。它是 xAI 的终端 Coding Agent，可以读代码、改文件、跑命令，还支持 Skills、MCP、插件和并行 Subagent。官方最新代码提交已经直接删除产品里的「Beta」标签。 但 1.0 并没有塞进什么重量级新功能。主要更新集中在稳定性和交互：大项目恢复不再容易卡死，后台任务取消后不会偷偷重启模型，连续发送消息不再丢队列，MCP 返回大截图不再损坏，大型会话占用的内存也明显收紧。Dashboard 还会直接显示 Agent 上一轮干了什么。 Grok Build 已经在 0.x 阶段狂更了近三个月，官网 Changelog 最早可以追到 0.1.209。 xAI
❤3
阿里把AI语音做成生产力套件，录音、Agent、播客全包 阿里上线一站式 AI 语音平台 CosyVoice Studio，底层使用 Qwen-Audio 系列模型，一口气包下语音记录、实时语音 Agent 和音频创作。 CosyFlow 主打语音记录。用户开口说话，它会自动去掉「嗯、啊、那个」等填充词和重复表达，还能把零散口述直接整理成清单、邮件和会议纪要。录音时还能区分不同发言人，自动生成摘要和待办。 CosyAgent 面向企业语音服务。用户用自然语言就能创建实时对话 Agent，还能接入企业文档、数据库、MCP 和 API，用于客服、电话营销等场景。 CosyCreative 则负责音频生产。上传文档、网页或输入文字，就能生成播客和多角色有声书，也支持声音复刻、局部重生成和多轨编辑。 个人端现已覆盖 iOS、Android、Mac 和 Windows，限时免费体验。CosyAgent 和 CosyCreativ…
👍1
前OpenAI研究员创业：把Codex换个皮，准备卖给所有白领 OpenAI 前 Sora 研究员 Gabriel Petersson 离职三个月后创业，推出办公 Agent Energy。 故事讲得很大，产品却很眼熟。它基本就是把 Codex、Cowork 那套 Agent 工作流搬到了白领办公。你说一句想干什么，它自己开浏览器、翻邮箱、找文件、点网页，最后把结果交回来。 相比 Cowork 和 ChatGPT Work，它主要多了自由切换模型，以及把记忆、Skills、自动化提前装成「一键助手」。有人直接问 Petersson，这是不是「另一个 Codex App」。他也承认，Cowork 和 ChatGPT 的方向是对的。 OpenAI 早在 6 月就开始把 Codex 往白领工作推，报告、表格、PPT、研究和工作流自动化都已经覆盖。Energy 首版暂时没拿出更强的新东西，只是把同一套玩法换了个壳。 这反倒替 …
🤡4
Agent插件终于要统一了，Vercel拉上OpenAI、微软、Cursor定标准 Vercel 联合 AWS、Cursor、GitHub、微软和 OpenAI 推出 Agent Plugins 1.0.0，给 AI Agent 插件定了一套开放的通用打包格式。过去同一个 Skill 或 MCP 服务，换到不同客户端往往还要重新整理目录和配置。现在按统一格式打包，兼容客户端就能直接识别和加载。 首批已经支持 ChatGPT、Codex、Cursor、GitHub Copilot、Kiro 和 VS Code。1.0 版只统一 Skills 和 MCP，命令、Hooks 等能力仍由各家自己实现，安装、权限和分发也不归它管。 Vercel
Showing the 12 most recent of 39 posts we hold for @OneMillion_AI. View and reaction counts are the latest single reading for each post, not a live figure, and a recent post is still accumulating both. A view count marked ≈ was rounded by Telegram before we ever saw it — t.me prints views in full below 1,000 and to three significant figures above, so ≈1,200,000 means somewhere between 1,150,000 and 1,249,999. Unmarked counts are exact. Text is reproduced from the public post preview and truncated for length.
Republished by
Channels on the register that have forwarded this channel's posts into their own feed.
Built only from forwarded posts we have actually read, on both sides. Coverage is early and deliberately incomplete: a missing link means we have not read the post that would prove it, never that the relationship does not exist. Counts are distinct forwarded posts observed, so they only ever go up as we read more.
Named by 5 registered channels — every channel on the register whose own posts have named this one, by its current username or any other username it currently holds, merged from two separately captured readings of the same fact so a namer caught by only one of them is not missed and a namer both caught is not counted twice. A username this channel has since dropped is not matched — that handle may belong to someone else now, and crediting today’s namer to yesterday’s owner would misattribute it.
Named by
Channels on the register whose posts name this channel's handle.
A mention is a weaker signal than a forward and is counted separately for that reason — naming a channel is not republishing it, and a handle in a post body is easy to place deliberately. The post counts beside each row below are distinct posts in which the handle appeared, from posts we have read on both sides — the “Named by N registered channels” figure above is a different count, of distinct NAMING CHANNELS rather than posts, and is not the sum of the rows under it.
A live page changes as we take new readings, so a citation should name the measurement it is based on, not just the URL. The line below cites the subscriber count as measured 27 September 2026 — this entry's latest reading, not the date you are reading this.
“动察Beating AI News” (@OneMillion_AI), 3,856 subscribers as measured 27 September 2026. Telegram Register, tgregister.com/channel/OneMillion_AI.
Full measurement history, CC BY 4.0. Every reading this register holds for this entry, not just the latest one, as a dated, downloadable record: CSV · JSON. Free to use with attribution to tgregister.com. Each file carries its own generation timestamp, which is the figure to cite for exactly when the data was retrieved.