[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-google-gemini-3-6-flash-lite-cyber":3,"news-related-5efc7b2d-a44c-4bfb-9a6c-5a8b39a35181":36},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":23,"news_slug":29,"published_at":30,"created_at":31,"modified_at":32,"is_published":33,"publish_type":34,"image_url":13,"view_count":35},"5efc7b2d-a44c-4bfb-9a6c-5a8b39a35181","谷歌三连发 Gemini 3.6 Flash \u002F Flash-Lite \u002F Flash Cyber:把 Agent 成本往下砍","7 月 21 日,Google 一次性放出 Gemini 3.6 Flash、3.5 Flash-Lite 和 3.5 Flash Cyber 三款新模型。和 3.5 Pro 迟迟不露面的谨慎相比,Google 在 Flash 层级的更新节奏明显加快——三版本同天上线,目标明确:高吞吐、低延迟、可大规模复制的 Agent 工作流。 3.6 Flash 是这次的主力。按 Artificial Analysis Index,质量不降的同时输出 token 较 3.5 Flash 砍 17%,DeepSWE 最高省 65%,定价 1.50\u002F7.50 美元每百万 token。Computer Use 作为客户端内置工具原生开放,OSWorld-Verified 从 78.4 提到 83.0。博客里直接亮出 DeepSWE 49% vs 37%、MLE Bench 63.9% vs 49.7% 等对比,卖点很清晰:质量没掉,单任务成本更低。 3.5 Flash-Lite 把\"便宜大碗\"做到极致:350 tokens\u002Fs 输出、0.30\u002F2.50 美元每百万 token。对标自家上一代 3 Flash,SWE-Bench Pro(54.2% vs 49.6%)和 OSWorld-Verified(74.0% vs 65.1%)都已反超。Google 把\"高 QPS、文档批量处理\"这类流水线任务彻底独立成档,主力模型不必再硬扛。 3.5 Flash Cyber 是另一信号:网络安全专用,仅向政府和受信任伙伴开放,在 CodeMender 框架内多 Agent 协同跑 CyberGym。同一周 OpenAI 刚爆出 GPT-5.6 Sol 红队测试中失控牵连 Hugging Face——头部厂商一边收紧攻击端能力、一边加速给防御侧配专用模型,分化下半年大概率还会继续放大。 3.5 Pro 仍只一句\"正在合作伙伴测试\",Gemini 4 预训练却已启动。Google 策略清晰:Fash 家族先把\"能上生产、能算账\"的盘子稳住,Pro 和下一代留给更大的叙事窗口。","https:\u002F\u002Fblog.google\u002Finnovation-and-ai\u002Fmodels-and-research\u002Fgemini-models\u002Fgemini-3-6-flash-3-5-flash-lite-3-5-flash-cyber\u002F","35ce748f-48b7-4638-88ef-effa57a7e749",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"7ac06d8e-b074-4147-abfc-ffaa4c6b8744","ai-efficiency",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"a9524a82-a7c5-4daa-bb4b-a7ee77bb0b94","gemini",{"id":18,"name":19,"slug":19,"description":13,"color":13},"8cf7490f-2449-4ba7-be19-61befa0d92b4","google",{"id":21,"name":22,"slug":22,"description":13,"color":13},"7e89b5cc-57db-4f37-bc6d-28919a73931c","model-release",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":28},"25d4c9d6-ba1c-403f-9c14-e44c9fd5fc3c","en","Google's triple Gemini 3.6 Flash launch cuts agent costs","On July 21, Google released three new models in one go: Gemini 3.6 Flash, 3.5 Flash-Lite and 3.5 Flash Cyber. Compared to 3.5 Pro's continued absence, Google's update cadence at the Flash tier is visibly accelerating — three versions shipping the same day, with a clear goal: high-throughput, low-latency, large-scale-replicable Agent workflows. 3.6 Flash is the headliner. Per the Artificial Analysis Index, output tokens drop 17% versus 3.5 Flash at equal quality, and DeepSWE saves up to 65%; pricing sits at $1.50 \u002F $7.50 per million tokens. Computer Use is opened natively as a built-in client-side tool, and OSWorld-Verified jumps from 78.4 to 83.0. The blog directly posts the DeepSWE 49% vs 37% and MLE-Bench 63.9% vs 49.7% comparisons — the pitch is crystal clear: quality unchanged, per-task cost lower. 3.5 Flash-Lite pushes the \"cheap and large\" to the extreme: 350 tokens\u002Fs output, $0.30 \u002F $2.50 per million tokens. Against the previous-generation 3 Flash, SWE-Bench Pro (54.2% vs 49.6%) and OSWorld-Verified (74.0% vs 65.1%) are already passed. Google has fully carved out \"high-QPS, document batch processing\" pipeline tasks into a separate tier, so the flagship model no longer has to bear everything. 3.5 Flash Cyber is another signal: cybersecurity-specialized, available only to governments and trusted partners, running multi-Agent collaboration in the CodeMender framework on CyberGym. The same week OpenAI just disclosed GPT-5.6 Sol losing control in red-team testing and dragging Hugging Face in — top vendors are simultaneously tightening offensive-side capabilities and accelerating dedicated models for the defensive side; the split will probably continue to widen in the second half. 3.5 Pro is still just \"in partner testing\", but Gemini 4 pretraining has already started. Google's strategy is clear: let the Flash family stabilize the \"production-ready, cost-accountable\" plate first; leave Pro and the next generation to a bigger narrative window.","google-gemini-3-6-flash-lite-cyber","2026-07-22T02:02:00Z","2026-07-22T02:07:02.939696Z","2026-08-19T02:08:40.142862Z",true,"agent",84,{"items":37},[38,43,48,53,58,63],{"id":39,"title":40,"news_slug":41,"published_at":42},"16856034-439d-4915-aed4-80b42ae09c68","Gemini 3.7 Flash：FrontierCode 43.6%，价格腰斩","gemini-3-7-flash-coding-agent-fast","2026-08-13T09:00:00+00:00",{"id":44,"title":45,"news_slug":46,"published_at":47},"5ef6f8fe-9877-4632-bb4a-690c7e73975e","Gemini 3.5 Flash 内置 Computer Use：OSWorld 78.4 把屏幕操控推成工程能力","gemini-3-5-flash-computer-use-osworld-78","2026-06-26T00:00:00+00:00",{"id":49,"title":50,"news_slug":51,"published_at":52},"e4497b0e-8295-46e8-b395-5f29a19ff26c","Gemini 3.5 Live Translate：当语音翻译告别「回合制」","gemini-3-5-live-translate-streaming","2026-06-10T00:00:00+00:00",{"id":54,"title":55,"news_slug":56,"published_at":57},"b8982a60-0c9f-4e3b-93b6-65828f938006","Gemini 3.5 Flash 重新定义「快」与「强」：Dynamic Thinking 如何打破 AI 推理的不可能三角？","gemini-3-5-flash-dynamic-thinking-4x","2026-05-28T01:00:00+00:00",{"id":59,"title":60,"news_slug":61,"published_at":62},"3481c485-e5ec-4a5d-9d4d-8fa92c845779","Google 发布 Gemini 3.5 Flash：面向 Agent 时代的编程与推理新旗舰","gemini-3-5-flash-agent-coding-4x-fast","2026-05-19T22:00:00+00:00",{"id":64,"title":65,"news_slug":66,"published_at":67},"386ce7fe-6fde-4d4a-8438-8b90f16bb963","Gemini 3.1 Flash-Lite 正式版发布：Google 最快最便宜的 Gemini 3 模型来了","gemini-3-1-flash-lite-ga-multimodal-1-50","2026-05-08T11:04:00+00:00"]