大逃杀

1.5/7.5 美元、便宜 17%：谷歌一天发三个 Gemini，却把最贵的旗舰藏着没发_我的网站

生死狙击

A | 谷歌干了一件很"谷歌"的事：不声不响，一口气把三个模型全发了。

A visitor listens to dialects of various regions in Hunan at the Exhibition Hall of the Project for the Protection of Language Resources of China in Changsha, central China's Hunan Province, Aug. 24, 2026. A new exhibition hall dedicated to the preservation of China's diverse language resources opened on Monday in Changsha, capital of central China's Hunan Province, bringing a decade-long effort to document languages and dialects to the public through digital and interactive displays. The Exhibition Hall of the Project for the Protection of Language Resources of China, located in the Hunan Museum, brings together recordings and other language materials collected from more than 1,800 field survey sites across China, covering Chinese dialects, ethnic minority languages and related cultural traditions. (Xinhua/Xue Yuge)Putting on headphones at a new exhibition hall in central China, visitors can hear a chorus of voices from across the country, including regional dialects, traditional songs in ethnic minority languages and field recordings of languages on the verge of extinction.
The Exhibition Hall of the Project for the Protection of Language Resources of China opened Monday in Changsha, capital of central China's Hunan Province, bringing decade-long efforts to document languages and dialects to the public through digital and interactive displays.
Located in the Hunan Museum, the exhibition brings together recordings and other language materials collected from more than 1,800 field survey sites across China. More than 30 language-related cultural relics are also on display.
The recordings represent 10 major Chinese dialects and more than 100 ethnic minority languages, including Mongolian, Tibetan, Uygur and Zhuang, said Chen Xuliang, deputy Party secretary of the Hunan Museum.
"Digital technology is helping bring the results of language preservation research out of databases and into public cultural spaces," he said.
The exhibition draws on materials collected through a national language preservation project launched in 2015 to document and preserve the country's diverse languages and dialects. It is the world's largest project of its kind, according to experts.
Over the past decade, researchers have traveled to remote mountain villages and pastoral areas to find native speakers of different ages. They systematically recorded words, grammatical examples, oral traditions and other language materials, turning oral memories at risk of disappearing in digital records that can be preserved for generations.
The work has become increasingly urgent as the number of native speakers of some languages declines.
China is home to more than 100 languages and numerous dialects. Some languages, including Hezhen and Manchu, now have fewer than 100 speakers.
Economic integration, population mobility and the social changes they bring are making it increasingly difficult to preserve languages and cultural traditions around the world, said Wang Lining, a professor at Beijing Normal University's School of Chinese Language and Literature and a participant in the language preservation project.
Many languages lack well-developed writing systems, Wang said, meaning knowledge and memories passed down orally can be lost when their speakers die.
The new exhibition aims to bring such preservation work closer to the public. It also promotes the principles of the "Yuelu Proclamation," which was officially released in 2019 and described as UNESCO's first permanent document themed on the protection of linguistic diversity.
Experts attending the exhibition's opening said a key challenge is how to use artificial intelligence and digital museum technologies to turn vast collections of field recordings into cultural experiences the public can engage with.
For Wang, documenting endangered languages is only the beginning.
"In the past, the focus was on recording endangered languages and preserving samples of their voices," she said. "In the future, preservation needs to move toward keeping languages alive."
The exhibition hall, Wang said, is part of that shift, taking language resources beyond academic research and bringing them closer to everyday life.

People visit the Exhibition Hall of the Project for the Protection of Language Resources of China in Changsha, central China's Hunan Province, Aug. 24, 2026. A new exhibition hall dedicated to the preservation of China's diverse language resources opened on Monday in Changsha, capital of central China's Hunan Province, bringing a decade-long effort to document languages and dialects to the public through digital and interactive displays. The Exhibition Hall of the Project for the Protection of Language Resources of China, located in the Hunan Museum, brings together recordings and other language materials collected from more than 1,800 field survey sites across China, covering Chinese dialects, ethnic minority languages and related cultural traditions. (Xinhua/Xue Yuge)

This photo taken on Aug. 24, 2026 shows an interior view of the Exhibition Hall of the Project for the Protection of Language Resources of China in Changsha, central China's Hunan Province. A new exhibition hall dedicated to the preservation of China's diverse language resources opened on Monday in Changsha, capital of central China's Hunan Province, bringing a decade-long effort to document languages and dialects to the public through digital and interactive displays. The Exhibition Hall of the Project for the Protection of Language Resources of China, located in the Hunan Museum, brings together recordings and other language materials collected from more than 1,800 field survey sites across China, covering Chinese dialects, ethnic minority languages and related cultural traditions. (Xinhua/Xue Yuge)

Children perform at the opening ceremony of the Exhibition Hall of the Project for the Protection of Language Resources of China in Changsha, central China's Hunan Province, Aug. 24, 2026. A new exhibition hall dedicated to the preservation of China's diverse language resources opened on Monday in Changsha, capital of central China's Hunan Province, bringing a decade-long effort to document languages and dialects to the public through digital and interactive displays. The Exhibition Hall of the Project for the Protection of Language Resources of China, located in the Hunan Museum, brings together recordings and other language materials collected from more than 1,800 field survey sites across China, covering Chinese dialects, ethnic minority languages and related cultural traditions. (Xinhua/Xue Yuge)
。不是挤牙膏式的"我们升级了一下"，是 Gemini 3.6 Flash、Gemini 3.5 Flash-Lite、Gemini 3.5 Flash Cyber 三款同时落地。你要是只看标题，会觉得"哦，又发模型了"。但你要是把这三款摊开看，会发现谷歌这次下的不是产品棋，是战略棋。它真正想说的，恰恰是一个它还没发出来的东西。先说最主力的那个，3.6 Flash。定价是输入 1.5 美元 / 百万 token、输出 7.5 美元 / 百万 token，知识截止 2026 年 3 月。

B | 光看价格没感觉，关键在后面那句：输出价格比前代降了大约 17%。

C | 更狠的是，它把 Computer Use（让模型直接操作电脑）做进了模型本身，还拿到了 AA Index 50 的评测分。翻译一下：这不是一个"陪你聊天的模型"，这是一个"能替你点鼠标、跑流程"的智能体底座，而且便宜了快两成。那 3.5 Flash-Lite 是什么角色？一个字：便宜。输入 0.3 美元、输出 2.5 美元每百万 token，是整条产品线里最便宜的那个，主打高吞吐、低成本场景。说白了，它是给那些"量很大、但每件事不复杂"的任务准备的——批量分类、简单改写、海量预处理。谷歌把"贵"和"便宜"两条线都铺好，让开发者几乎没理由不用它家的 Flash 系列。

D | 最让我意外的是第三个，3.5 Flash Cyber。

E | 这是个网络安全专用模型，搭配 CodeMender 代码安全智能体，干的是漏洞识别和修复的活儿。而且它不开放公开 API，只以受限试点形式给政府机构和"信任伙伴"用。你品品这个信号：谷歌把"AI 做安全"单独切出来，做成一款只对特定对象开放的产品。这不是技术秀，这是把 AI 安全变成一门To G 的生意。把这三款摆到一起，你就能看出它在对标谁。3.6 Flash 自带 Computer Use，摆明了是冲着"能替人干活的 AI 员工"市场去的——那块原本是 OpenAI Codex 和 Claude 的腹地。Flash-Lite 用 0.3 美元的地板价，直接压 OpenAI 和 Anthropic 的企业 API 定价。

F | 而那个只对政府开放的 Cyber，则是在抢 Palo Alto、CrowdStrike 这些老牌安全厂商的饭碗。谷歌这一手，是用一条产品线，同时捅了三个赛道的腰眼。这里面有个更深的算盘。谷歌敢把旗舰藏着、把 Flash 铺满，本质是赌一件事：未来的模型竞争，不再是"谁更聪明"，而是"谁更便宜、更能嵌进工作流"。当算力成本随时间往下走、模型能力随规模往上走，"够用且便宜"会吃掉"极致聪明"的大部分市场。这跟当年 x86 把一众精美架构按在地上摩擦是同一个逻辑——不是你更强，是你更无处不在。谷歌现在要当的那个"无处不在"，就是 Flash。但真正吓人的，是谷歌没发的那个。大家都在等的旗舰 Gemini 3.5 Pro，到今天还躺在"合作伙伴测试"里，连个确切日期都没有。而就在同一天，谷歌甩出一句话：已经启动了公司史上"最雄心勃勃的预训练"——也就是 Gemini 4。你看明白没？它一边用 Flash 系列把中低端市场和智能体场景占满，一边把旗舰的发布往后拖，直接去押下一代的"世代跨越"。这招很毒：用成熟产品赚着钱、占着生态，同时不让对手有"旗舰对标"的靶子。我得说句公道话：3.5 Pro 一直没发，未必是"难产"，更可能是主动选择。当 3.6 Flash 已经把"聪明够用 + 便宜 17% + 能操作电脑"这三件事打包好，旗舰的紧迫感就被稀释了。用户要的是"能用、不贵、能干活"，不一定非要"最聪明那个"。谷歌赌的就是这一点。把这三款连起来看，谷歌的打法就清晰了：它正在用一套价格梯度 + 场景梯度的组合拳，把 AI 模型的"性价比"重新定义。贵的旗舰我先藏着，便宜的 Flash 我先铺满；能跑流程的给开发者，能搞安全的给政府。对手还在比"谁的参数多、谁的榜单高"，谷歌已经改成比"谁的生态卡得死、谁的客户跑不掉"。对咱们普通开发者来说，这其实是好事。3.6 Flash 便宜了 17% 还自带 Computer Use，意味着"用 AI 跑一个自动化流程"的成本门槛又低了一截。原来你舍不得调用的智能体，现在可以放心跑了。对创业公司更明显——Flash-Lite 这种廉价高吞吐模型，简直是为"用 AI 批量处理业务"量身定做的。但我也要泼盆冷水。便宜是便宜，可别忘了：你把业务越往 Flash 上搬，就越被谷歌的云服务绑死。今天它降你 17%，明天它涨你 20%，你迁移的成本会比省下的那点 token 钱高得多。模型可以随便换，工作流一旦长在别人平台上，就不好拔了。

G | 还有件事得说透：所谓"输出价格降 17%"，省的是 token 钱，不是你的时间钱。模型跑得再便宜，如果你的业务流程没 redesign 成"适合智能体跑"的样子，那 17% 也就是账面上好看。真正吃到这波红利的，是那些早就把任务拆成"能并行、能自动化"的团队。

H | 红利从来不是均匀落的，它只落给早做准备的人。

I | 所以这次三个模型齐发，表面是"产品更新"，底子是"生态卡位"。谷歌想抢的，不是你下一次的对话，而是你未来三年跑在它上面的那整套业务流程。旗舰 3.5 Pro 不发也好，发也罢，那张棋盘上真正要赢的，从来不是单颗棋子的强弱。对我们来说，眼下最实在的一句提醒是：趁它还便宜，赶紧把该自动化的跑起来；但别把命根子也一起交上去。

J | 用 Flash 跑流程很爽，记得在每个关键环节留一个"能随时迁走"的备份方案。免费的便宜，从来都标着价，只是账单晚一点到。

Current article:http://u4k4j1.qingfanceliangsangniang.cfd/news/20260826_99029.html

Published on:04:52:52