全网AI产品免费权益清单 云知声U2-Flash智谱AutoClaw大额度Token到期日汇总Roundup of Free AI Product Perks: Cloudminds U2-Flash, Zhipu AutoClaw and Large-Token Quota Expiry Dates


AI free token/AI免费积分

全网AI产品免费权益清单 云知声U2-Flash智谱AutoClaw大额度Token到期日汇总

本文来源:https://post.smzdm.com/p/ae65l8m3
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

这两天国内 AI 圈的主基调是"一边开源、一边收口"。 腾讯 把 Hy4 flash 的三版权重放了出去,本地部署又多一个能打的选项;同时有几处免费口子在悄悄关上,腾讯云把 DeepSeek 联网搜索开始计费、MiniMax M3.1 免费版从 OpenCode 池子里下了架。下面按"新冒出来的 / 正在关窗的 / 还能慢慢领的"三段整理,每条都标了日期和入口,先领先用、别囤。

腾讯混元 Hy4 flash 三版权重开源,本地部署不收授权费

11 月 17 日至 18 日,腾讯混元把 Hy4 flash 的三个版本——标准指令版、FP8 量化版、Base 预训练版——放到 GitHub、Hugging Face、ModelScope、GitCode 上,权重免费下载。模型 220B 总参、每 token 激活约 12B,原生多模态输入,上下文 100 万 token,官方给了 vLLM / SGLang 镜像。企业把权重拉到内网做文档、PPT、前端辅助这类日常用途,不用额外交授权费。

参与方式:到上述四个平台的腾讯混元官方仓库搜 "Hy4 flash",下载对应版本;FP8 版适合生产部署,Base 版适合继续训练。

截止时间:无,属开源权重、长期可下,注意这不是 API 免费,自己部署的电费 显卡 钱照旧。

出处: 新浪财经·腾讯 Hy4 flash 开源,企业内部部署需要收授权费吗 、 Pandaily·Tencent Releases Downloadable Hy4 flash Weights in Three Variants

解读: "权重开源"和"你能随便商用"是两回事。它用的是腾讯 Hy License 1.0(不是标准 Apache 2.0),内部使用明确豁免,但如果你拿它对外提供模型服务、年营收超 5000 万美元或产品月活过 1 亿,就得单独找腾讯拿书面授权。中小企业和自建内网基本随便用,做 API 生意的先看协议第 10 条。

腾讯云 TokenHub:DeepSeek 联网搜索、网页抓取从免费转收费

11 月 18 日零点起,腾讯云 TokenHub 里 DeepSeek 原厂直供模型的联网搜索(Search)和网页抓取(Fetch)开始计入 Token Plan 积分抵扣,各按 100 积分/千次成功请求算,折合约 5 元/千次,跟 DeepSeek 官方同价;搜索结果注入上下文的部分还要再按输入扣一次积分。此前走这条通道做联网问答的套餐用户,这几天账单会多一块。

解读: 这条不是新福利,是提醒你"免费的少了"。免费侧的替代口子还在—— 阿里云 联网搜索 2000 次到 11-30 才升级、火山联网资源插件每月 2 万次、火山 Managed Agents 的 web_fetch 目前还限免(见下一条)。做联网问答的把这三条对比一下再决定走哪条。

火山方舟 Managed Agents:web_fetch 限时免费,首开另赠 500 次

11 月 18 日,火山把 Managed Agents 的计费挂上了价格页:web_search 0.02 元/次、web_fetch 限时免费、Agent 运行时 0.5 元/小时。首次开通 web_fetch 另送 500 次。

参与方式:火山方舟控制台进 Managed Agents,开通即可用 web_fetch 免费额度。

截止时间:web_fetch 写的是"限时免费",官方没给硬日期。

OpenCode「GLM 行动」补贴翻倍:付 10 美元享 200 美元 GLM-5.3 用量

11 月 18 日,OpenCode 把自家 GLM 行动第二阶段补贴从"付 10 美元得 100 美元"翻到"付 10 美元得 200 美元"GLM-5.3 用量,仅限 Go 订阅用户。GLM-5.3 官方定价没变。

截止时间:公告没写截止,也没说是一次性额度还是订阅内权益。

解读: 这条是折扣、不是白送,掏的是 10 美元真金白银。折算下来确实是这块表里单位成本很低的一档,但注意边界——等于别人替你付了那 190 美元,补贴一停就没了,别当长期饭票。

6 款新模型免部署 0 元通道,网页和桌面两条路都整理好了

11 月 18 日有人把当前在架的免费新模型按入口归了一遍,两条零元路线:

网页直聊(OpenRouter 搜 ":free"):百度 ERNIE-5.5-Flash:free,280B 总参/14B 激活、100 万级上下文、千帆官方节点直供、实测约 86 字/秒;蚂蚁 Ling-3.2-Flash:free,560B 总参、单次最多传 200 张图、可开关思考。没充值的纯免费账号全站共用约 20 次/分、50 次/天。

桌面干活(OpenCode 桌面端 Zen 池,零数据留存):Hy4 flash Free、DeepSeek V4.1 Flash Free、美团 LongCat 2.5 Free(1.6 万亿总参、国内托管、每天 100 次)、GLM-5.3-Flash Free、小米 MiMo-V2.6-Flash Free。

另有 Inkling 2 Small:free(前 OpenAI CTO 团队)只在 Agent 编程工具里能调,网页聊天窗会拦,且官方明确会用你的提问和输出训练模型,别传人脸、证件、公司内部代码。

截止时间:多为厂商攒口碑的限免窗口,11-18 核对在架,随时可能下。

出处: 微信公众号·6 款新模型免费指南(11-18 核对)

蚂蚁百灵 Ling-3.2-flash:今天最后窗口

上线前两周对话和 API 全免,官方只说"前两周"、没给硬日期,按 11 月初发布倒推,窗口落在 11 月 19 日至 20 日。要用的按今天 11 月 19 日做准备更稳,别赌 20 号。

出处: All Agent 百科·AI 免费额度 11 月 17 日 、 AI 模型优惠速报 11-19 档

火山 Coding Plan 邀请 9.5 折:今天最后一天

被邀请人首单 9.5 折、邀请人得 5% 代金券,11 月 19 日为最后一天。有邀请链路的今天用掉。

MiniMax M3.1 免费版:已在 OpenCode 池下线

11 月 17 日,OpenCode 免费池里的 MiniMax M3.1 免费版到点下线,照旧清单找会扑空。

云知声 U2-Flash :1 亿 Token 免费档,11-20 到期。

OpenCode 限时免费模型池 :M3.1 Flash、V4.1 Flash、GLM-5.3-Flash 等多款仅 OpenCode 内免费用,LongCat 每天 100 次,11-23 关池。

阿里 Qwen4-Max 预览版 :预览期 2 折权益到 11-25 24:00,qwen4-max-preview 标识当天正式下线(正式版不受影响);不花钱试可走 QwenStudio 网页版。

腾讯混元 Hy4 flash :WorkBuddy(了解更多) 客户端内 0.00x 免费用到 11-30 23:59(每日送免费额度、用完可用积分续);TokenHub 的 API 五折同日到 11-30 12:00。

无问 AI 兑换码 WUWEN_DEC_TOKEN_TICKET :兑 4500 万限期包,12-09 11:51 清零,注册后每天另到账 3500 万日额度。

长期白拿、没变 :阿里百炼 7000 万免费 Token(90 天)、Qoder 系数 0 不限量、智谱 ZCode 每日 3 亿夜间档 + Start Plan 每日 800 万连送 5 天、商汤 Token Plan 公测 ¥0/月、国家超算 AIGC 短视频四选一限免(11-30)、讯飞 SparkAudio 认证免费包(个人 200 万/企业 500 万、1 年)、腾讯 Hy-SimulMT 开源自部署、腾讯云 TokenHub 每模型 100 万(12-31)。

出处: AutoClaw 官方博客 、 阿里云百炼官方文档 、 AI 模型优惠速报 11-19 档 、 什么值得买·11-18 清单

解读: 这一段都是老面孔、没有大变动,纯粹替你记着日子。原则还是那句"有效期比额度重要"——19、20、22、23、25、30 这几个节点跟着的福利,先领先用、别囤着等过期。

通用的 AI 兑换码 / CDKey 还是没有,无问那个是官方月度活动、不属于通用码,别拿去别的平台乱试。折扣也不等于免费,OpenCode GLM 行动、Qwen4-Max 2 折、GLM 行动翻倍这些都是付费套餐里打折,掏钱之前先看清门槛。

权重开源不等于无条件商用——Hy4 flash 的 Hy License 对大规模对外服务划了线,做 API 生意的先看协议。低价包绑了支付方式记得关自动续费;领之前认准官方入口,地方词元券不是全国通用。羊毛有时效,以你打开客户端那一刻显示的为准。

Roundup of Free AI Product Perks: Cloudminds U2-Flash, Zhipu AutoClaw and Large-Token Quota Expiry Dates

本文来源:https://post.smzdm.com/p/ae65l8m3
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

The main theme in China's AI community over the past couple of days has been "open-sourcing on one hand, and tightening up on the other." Tencent has released the three weights of Hy4 flash, adding another viable option for local deployment. Meanwhile, several free access points are quietly being closed. Tencent Cloud has started charging for DeepSeek's online search, and the free version of MiniMax M3.1 has been removed from the OpenCode pool. Below, I've organized everything into three sections: "Newly available / Closing soon / Still available for a limited time." Each item is marked with a date and access link. Use them as soon as you can—don't hoard them.

Tencent Hunyuan Hy4 flash three-license weight open-sourced; no licensing fees for local deployment

From November 17 to 18, Tencent Hunyuan made three versions of Hy4 flash — the Standard Instruction Version, the FP8 Quantized Version, and the Base Pre-trained Version — available on GitHub, Hugging Face, ModelScope, and GitCode, with the weights available for free download. The model has 220B total parameters, approximately 12B activations per token, native multimodal input, and a context of 1 million tokens. The official vLLM/SGLang image is provided. Companies can pull the weights into their intranet for everyday uses such as document creation, PPTs, and front-end assistance without paying additional licensing fees.

How to participate: Search for "Hy4 flash" in the official Tencent Hunyuan repositories on the four platforms mentioned above and download the corresponding version. The FP8 version is suitable for production deployment, while the Base version is suitable for continued training.

Deadline: None. These are open-source weights and can be downloaded long-term. Please note that this is not a free API; you will still need to pay for electricity and graphics cards for self-deployment.

Sources: Sina Finance • Tencent Hy4 flash open source – Is a licensing fee required for internal enterprise deployment? • Pandaily • Tencent Releases Downloadable Hy4 flash Weights in Three Variants

Interpretation: "Open-source weights" and "you can use them for commercial purposes as you wish" are two different things. It uses the Tencent Hy License 1.0 (not the standard Apache 2.0), with an explicit exemption for internal use. However, if you use it to provide model services externally, have annual revenue exceeding $50 million, or have a product with over 100 million monthly active users, you must obtain separate written authorization from Tencent. Small and medium-sized enterprises and companies with self-built intranets can generally use it freely. Those in the API business should first refer to Article 10 of the agreement.

Tencent Cloud TokenHub: DeepSeek's online search and web scraping are transitioning from free to paid.

Starting at midnight on November 18, the online search (Search) and web scraping (Fetch) features of the DeepSeek models directly provided by the original manufacturer in Tencent Cloud's TokenHub will start to be counted towards Token Plan point deductions. Each is calculated at 100 points per thousand successful requests, which is approximately 5 yuan per thousand requests, the same price as on the official DeepSeek platform. Additionally, points will be deducted once more based on the input for the part where search results are injected into the context. Package users who previously used this channel for online Q&A will see an additional charge on their bills over the next few days.

Explanation: This is not a new benefit, but a reminder that "there are fewer free options." Alternative free options still exist – Alibaba Cloud's online search will only be upgraded after 2,000 requests, until November 30; Volcano's online resource plugin offers 20,000 requests per month; and Volcano Managed Agents' web_fetch is currently still free for a limited time (see next item). If you're doing online Q&A, compare these three options before deciding which one to use.

Volcano Ark Managed Agents: web_fetch is free for a limited time, with an additional 500 requests for first-time users.

On November 18, Volcano posted the pricing for Managed Agents on the pricing page: web_search at ¥0.02 per request, web_fetch free for a limited time, and Agent runtime at ¥0.5 per hour. An additional 500 uses are provided when web_fetch is activated for the first time.

How to participate: Go to Managed Agents in the Volcano Engine console and activate to use the web_fetch free quota.

Deadline: web_fetch is listed as "limited-time free," and the official site hasn't provided a specific end date.

OpenCode "GLM Action" subsidy doubled: Pay $10 to enjoy $200 worth of GLM-5.3 usage

On November 18, OpenCode doubled its GLM Action Phase 2 subsidy from "$10 for $100" to "$10 for $200" worth of GLM-5.3 usage, exclusively for Go subscribers. The official pricing for GLM-5.3 remains unchanged.

Deadline: The announcement does not specify a deadline, nor does it indicate whether it is a one-time credit or a subscription benefit.

Interpretation: This is a discount, not a freebie. You're actually paying $10 in real money. When calculated, it really is one of the lowest unit cost options in this table, but be mindful of the limits – it's like someone else paying that $190 for you. Once the subsidy ends, it's gone. Don't treat it as a long-term freebie.

6 new models with no deployment required and a free access channel. Both web and desktop routes have been set up.

On November 18, someone categorized the currently available free new models by access point, creating two free access routes:

Web direct chat (search ":free" on OpenRouter): Baidu ERNIE-5.5-Flash:free, 280B total parameters/14B active parameters, 1 million-level context, direct access via Qianfan's official nodes, tested at approximately 86 characters/second; Ant Ling-3.2-Flash:free, 560B total parameters, can upload up to 200 images per request, with toggleable reasoning. Purely free accounts without top-ups share approximately 20 requests per minute and 50 requests per day across the entire platform.

Desktop options (OpenCode desktop Zen pool, zero data retention): Hy4 Flash Free, DeepSeek V4.1 Flash Free, Meituan LongCat 2.5 Free (1.6 trillion parameters, domestically hosted, 100 uses per day), GLM-5.3-Flash Free, Xiaomi MiMo-V2.6-Flash Free.

There is also Inkling 2 Small:free (from the former OpenAI CTO team), which can only be accessed through the Agent programming tool. The web chat window will block it, and the official team explicitly states that they will use your questions and outputs to train the model. Do not upload facial images, identification documents, or internal company code.

Deadline: Most of these are limited-time free access periods for vendors to build their reputation. As of November 18, they are still available, but they could be withdrawn at any time.

Source: WeChat Official Account – Free Guide to 6 New Models (verified as of November 18)

Ant Group's Ling-3.2-flash: Last chance today

Conversations and APIs are completely free for the first two weeks before launch. The official announcement only mentions "the first two weeks" without specifying a fixed date. Working backward from the early November launch, the window falls between November 19 and 20. If you plan to use it, it's safer to prepare for today, November 19th. Don't gamble on the 20th.

Source: All Agent Encyclopedia · AI Free Allowance November 17, AI Model Discount Update November 19 Edition

Volcano Coding Plan Invitation 5% Discount: Today is the Last Day

Invitees get a 5% discount on their first order, and inviters receive a 5% coupon. November 19 is the last day. If you have an invitation link, use it today.

MiniMax M3.1 Free Version: Has been removed from the OpenCode pool

On November 17, the MiniMax M3.1 Free Version in the OpenCode free pool was removed at the scheduled time. If you look for it in the usual list, you won't find it.

UniUni U2-Flash: 100 million tokens free tier, expires on November 20.

OpenCode Limited-Time Free Model Pool: Multiple models such as M3.1 Flash, V4.1 Flash, and GLM-5.3-Flash are available for free use only within OpenCode. LongCat is available 100 times per day. The pool will be closed on November 23.

Alibaba Qwen4-Max Preview Version: 80% off during the preview period until 11/25 at midnight. The qwen4-max-preview identifier will be officially retired on the same day (the official version will not be affected). You can try it for free on the QwenStudio web version.

Tencent Hunyuan Hy4 Flash: 0.00x free usage in the WorkBuddy(了解更多) client until 11/30 23:59 (daily free quota provided, and available points can be used to top up once the quota is used up); TokenHub's API is 50% off until 11/30 12:00 on the same day.

Wuwen AI redemption code WUWEN_DEC_TOKEN_TICKET: Redeem for a limited-time package of 45 million tokens, which will be reset to zero on December 9 at 11:51 AM. After registration, an additional daily quota of 35 million tokens will be credited each day.

Long-term freebies, no changes: Alibaba Bailian 70 million free tokens (90 days), Qoder coefficient 0 unlimited, Zhipu ZCode 300 million daily night-time slot + Start Plan 8 million daily for 5 consecutive days, SenseTime Token Plan beta test ¥0/month, National Supercomputing AIGC short video choose one of four limited-time free offers (11-30), iFlytek SparkAudio certified free package (2 million for individuals/5 million for enterprises, 1 year), Tencent Hy-SimulMT open-source self-deployment, Tencent Cloud TokenHub 1 million per model (12-31).

Sources: AutoClaw Official Blog, Alibaba Cloud Bailian Official Documentation, AI Model Deals Update 11/19 Edition, What's Worth Buying – 11/18 List

Explanation: This section is full of familiar faces with no major changes; it's simply here to remind you of the dates. The principle remains the same: "Validity period is more important than the amount." For the benefits associated with the dates 19th, 20th, 22nd, 23rd, 25th, and 30th, use them first and don't hoard them until they expire.

There still aren't any universal AI redemption codes/CDKeys. Regardless of whether it's an official monthly event, it's not a universal code. Don't go trying it on other platforms randomly. Discounts don't equate to free either. The OpenCode GLM campaign, the Qwen4-Max 20% discount, and the GLM campaign double-up are all discounts on paid plans. Make sure you understand the eligibility criteria before spending any money.

Open-source weights do not equate to unconditional commercial use – the Hy License for Hy4 flash sets limits on large-scale external services. Those involved in API businesses should review the agreement first. If a low-price package is linked to a payment method, remember to turn off automatic renewal. Before claiming, make sure to use the official portal, as local token vouchers are not universally valid nationwide. Deals have an expiration date. The information displayed when you open the client is the most accurate.

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry