[{"data":1,"prerenderedAt":993},["ShallowReactive",2],{"header-counts":3,"footer-counts":6,"model-gemini-3-6-flash":9},{"tools":4,"reviews":5},78,26,{"tools":4,"reviews":5,"playbooks":7,"news":8},22,27,{"id":10,"title":11,"apiCompatible":12,"benchmarks":15,"body":16,"category":960,"contextWindow":961,"cover":15,"description":962,"extension":963,"lastVerified":15,"maxOutput":964,"meta":965,"navigation":287,"path":966,"pricing":967,"published":968,"relatedTools":969,"releaseDate":972,"seo":973,"slug":974,"stem":975,"strengths":976,"updated":968,"useCases":982,"vendor":987,"vendorEn":987,"weaknesses":988,"__hash__":992},"models\u002Fmodels\u002Fgemini-3-6-flash.md","Gemini 3.6 Flash",[13,14],"google","vertex",null,{"type":17,"value":18,"toc":934},"minimark",[19,23,48,51,73,79,82,87,94,102,116,119,126,151,154,157,176,179,182,202,205,208,212,215,219,222,226,235,239,247,251,254,367,371,512,515,578,581,633,639,653,656,659,759,765,800,803,835,838,864,868,874,880,886,892,895,930],[20,21,22],"h2",{"id":22},"概述",[24,25,26,27,31,32,35,36,39,40,43,44,47],"p",{},"Gemini 3.6 Flash 是 Google 于 ",[28,29,30],"strong",{},"2026 年 7 月 21 日","在 Gemini API 上线的",[28,33,34],{},"稳定版（stable）"," Flash 模型。它延续 Flash 系列的「低价、低延迟、原生多模态」定位，提供 ",[28,37,38],{},"1,048,576 token 输入上下文","与 ",[28,41,42],{},"65,536 token 输出上限","，官方定价 ",[28,45,46],{},"$1.50 输入 \u002F $7.50 输出 \u002F $0.15 缓存输入","（每百万 token）。",[24,49,50],{},"在 Gemini 的产品阶梯里，3.6 Flash 处在「性价比中枢」：",[52,53,54,61,67],"ul",{},[55,56,57,60],"li",{},[28,58,59],{},"Gemini 3 Pro \u002F 3.5 Pro","：能力天花板，贵，适合最难的任务；",[55,62,63,66],{},[28,64,65],{},"Gemini 3.6 Flash（本文）","：能力够用、便宜、快，绝大多数生产场景的默认档；",[55,68,69,72],{},[28,70,71],{},"Gemini 3.5 Flash-Lite","（同批上线，$0.30\u002F$2.50）：极致成本，适合超大规模、对质量容忍度高的自动化。",[24,74,75,76,78],{},"同批上线的还有 ",[28,77,71],{},"（7 月 21 日，定位超大规模自动化，$0.30\u002F$2.50 输入\u002F输出）。「stable」意味着它不再是预览\u002F实验通道，SLA、版本稳定性与定价都进入长期维护状态——这对生产接入很关键。",[20,80,81],{"id":81},"核心能力",[83,84,86],"h3",{"id":85},"原生多模态一次输入多种信号","原生多模态：一次输入，多种信号",[24,88,89,90,93],{},"Gemini 3.6 Flash 的图像、视频、音频与文本是",[28,91,92],{},"统一模态","（native multimodal），不需要像很多模型那样先把图片转成独立视觉 token 或走单独的视觉子模型。带来的直接好处：",[52,95,96,99],{},[55,97,98],{},"一条请求里可以同时塞「一段产品演示视频 + 几张截图 + 一段用户语音 + 文字指令」，模型一并理解；",[55,100,101],{},"适合「感知型」工作流：内容审核、视频摘要、带图客服、多模态 RAG。",[103,104,105],"blockquote",{},[24,106,107,108,111,112,115],{},"对比：不少模型对图片是「附加上下文」，Gemini 的强项是把视频\u002F音频也当成一等公民。如果你的场景只是「看图回答问题」，差异不明显；但若涉及",[28,109,110],{},"视频时序理解","或",[28,113,114],{},"音视频联合推理","，Flash 的多模态原生度更有体感。",[83,117,118],{"id":118},"长上下文工程实践",[24,120,121,122,125],{},"1M 输入上下文意味着你可以把",[28,123,124],{},"一整本书、一份几百页 PDF、或一个中型代码仓库","直接放进一次请求。但实际落地有讲究：",[52,127,128,134,140],{},[55,129,130,133],{},[28,131,132],{},"不要无脑塞满","：超长上下文单次成本更高，且模型对「中间段」信息的注意力会衰减（lost-in-the-middle）。优先放与当前任务最相关的片段。",[55,135,136,139],{},[28,137,138],{},"长文档 RAG 的新玩法","：对 1M 上下文，很多场景可以从「切块召回」退化为「整文档直读」，省掉向量库与召回链路，但要在成本与延迟间权衡。",[55,141,142,145,146,150],{},[28,143,144],{},"代码库问答","：把 ",[147,148,149],"code",{},"README"," + 关键模块 + 报错栈一起喂进去，比只贴报错栈有效得多。",[83,152,153],{"id":153},"低成本高吞吐",[24,155,156],{},"Flash 的存在意义就是「把前沿能力打到可批量的价格」。相比 Pro 系列：",[52,158,159,166,169],{},[55,160,161,162,165],{},"输入价只有 Pro 的零头，适合",[28,163,164],{},"高 QPS"," 场景（如每用户请求都调一次模型）；",[55,167,168],{},"延迟更低，交互式应用（客服、实时摘要）体验更好；",[55,170,171,172,175],{},"配合",[28,173,174],{},"输入缓存","（见定价小节），重复前缀能压到 $0.15\u002F1M。",[83,177,178],{"id":178},"函数调用与工具使用",[24,180,181],{},"Flash 原生支持：",[52,183,184,190,196],{},[55,185,186,189],{},[28,187,188],{},"Function Calling \u002F Tool Use","：模型可返回结构化工具调用，配合外部 API 编排；",[55,191,192,195],{},[28,193,194],{},"结构化输出 \u002F JSON 模式","：直接产出可解析的 JSON，省去后处理；",[55,197,198,201],{},[28,199,200],{},"流式（streaming）","：逐字返回，适合打字机式 UI。",[24,203,204],{},"这些是 Agent 流水线的地基——也解释了为什么它适合做「低成本 Agent 编排层」的底座模型。",[20,206,207],{"id":207},"适用场景详解",[83,209,211],{"id":210},"_1-多模态内容审核与标注","1. 多模态内容审核与标注",[24,213,214],{},"把用户上传的图片\u002F视频\u002F音频 + 平台规则文本一起发过去，让模型输出分类标签与理由。Flash 的低价让「每条 UGC 都过一遍模型」在经济上可行。",[83,216,218],{"id":217},"_2-长文档-大代码库理解","2. 长文档 \u002F 大代码库理解",[24,220,221],{},"合同、研报、技术文档的摘要与问答；或对仓库做「这段逻辑在干嘛 \u002F 哪里有重复 \u002F 这个报错怎么修」类提问。1M 上下文让「少切块」成为可能。",[83,223,225],{"id":224},"_3-低成本-agent-与自动化流水线","3. 低成本 Agent 与自动化流水线",[24,227,228,229,234],{},"把 Flash 作为多步骤 Agent 的「执行模型」：简单决策、信息抽取、格式转换由它完成，只在遇到真正难的子任务时升级到 Pro 或外部更强模型（参见 ",[230,231,233],"a",{"href":232},"\u002Fcoding\u002Fide\u002Fcursor.html","Cursor Router"," 这类请求级路由思路）。",[83,236,238],{"id":237},"_4-实时翻译摘要客服对话","4. 实时翻译、摘要、客服对话",[24,240,241,242,246],{},"低延迟 + 多语言 + 流式，天然适配客服与摘要。配合 ",[230,243,245],{"href":244},"\u002Fcoding\u002Fapi\u002Fopenrouter.html","OpenRouter"," 这类聚合层还能做多模型兜底。",[20,248,250],{"id":249},"api-调用示例","API 调用示例",[83,252,253],{"id":253},"基础文本",[255,256,261],"pre",{"className":257,"code":258,"language":259,"meta":260,"style":260},"language-python shiki shiki-themes github-light github-dark","from google import genai\n\nclient = genai.Client(api_key=\"YOUR_API_KEY\")\nresp = client.models.generate_content(\n    model=\"gemini-3.6-flash\",\n    contents=\"把这段产品评论按正面\u002F负面分类，并给出一句话总结\",\n)\nprint(resp.text)\n","python","",[147,262,263,282,289,314,325,339,352,357],{"__ignoreMap":260},[264,265,268,272,276,279],"span",{"class":266,"line":267},"line",1,[264,269,271],{"class":270},"szBVR","from",[264,273,275],{"class":274},"sVt8B"," google ",[264,277,278],{"class":270},"import",[264,280,281],{"class":274}," genai\n",[264,283,285],{"class":266,"line":284},2,[264,286,288],{"emptyLinePlaceholder":287},true,"\n",[264,290,292,295,298,301,305,307,311],{"class":266,"line":291},3,[264,293,294],{"class":274},"client ",[264,296,297],{"class":270},"=",[264,299,300],{"class":274}," genai.Client(",[264,302,304],{"class":303},"s4XuR","api_key",[264,306,297],{"class":270},[264,308,310],{"class":309},"sZZnC","\"YOUR_API_KEY\"",[264,312,313],{"class":274},")\n",[264,315,317,320,322],{"class":266,"line":316},4,[264,318,319],{"class":274},"resp ",[264,321,297],{"class":270},[264,323,324],{"class":274}," client.models.generate_content(\n",[264,326,328,331,333,336],{"class":266,"line":327},5,[264,329,330],{"class":303},"    model",[264,332,297],{"class":270},[264,334,335],{"class":309},"\"gemini-3.6-flash\"",[264,337,338],{"class":274},",\n",[264,340,342,345,347,350],{"class":266,"line":341},6,[264,343,344],{"class":303},"    contents",[264,346,297],{"class":270},[264,348,349],{"class":309},"\"把这段产品评论按正面\u002F负面分类，并给出一句话总结\"",[264,351,338],{"class":274},[264,353,355],{"class":266,"line":354},7,[264,356,313],{"class":274},[264,358,360,364],{"class":266,"line":359},8,[264,361,363],{"class":362},"sj4cs","print",[264,365,366],{"class":274},"(resp.text)\n",[83,368,370],{"id":369},"多模态输入图文联合","多模态输入（图文联合）",[255,372,374],{"className":257,"code":373,"language":259,"meta":260,"style":260},"from google import genai\nfrom google.genai import types\n\nclient = genai.Client(api_key=\"YOUR_API_KEY\")\nresp = client.models.generate_content(\n    model=\"gemini-3.6-flash\",\n    contents=[\n        \"根据这张截图和语音转写，判断用户遇到的具体问题并给排查步骤\",\n        types.Part.from_uri(\"gs:\u002F\u002Fyour-bucket\u002Fscreenshot.png\", mime_type=\"image\u002Fpng\"),\n        types.Part.from_uri(\"gs:\u002F\u002Fyour-bucket\u002Faudio.wav\", mime_type=\"audio\u002Fwav\"),\n    ],\n)\nprint(resp.text)\n",[147,375,376,386,398,402,418,426,436,445,452,475,494,500,505],{"__ignoreMap":260},[264,377,378,380,382,384],{"class":266,"line":267},[264,379,271],{"class":270},[264,381,275],{"class":274},[264,383,278],{"class":270},[264,385,281],{"class":274},[264,387,388,390,393,395],{"class":266,"line":284},[264,389,271],{"class":270},[264,391,392],{"class":274}," google.genai ",[264,394,278],{"class":270},[264,396,397],{"class":274}," types\n",[264,399,400],{"class":266,"line":291},[264,401,288],{"emptyLinePlaceholder":287},[264,403,404,406,408,410,412,414,416],{"class":266,"line":316},[264,405,294],{"class":274},[264,407,297],{"class":270},[264,409,300],{"class":274},[264,411,304],{"class":303},[264,413,297],{"class":270},[264,415,310],{"class":309},[264,417,313],{"class":274},[264,419,420,422,424],{"class":266,"line":327},[264,421,319],{"class":274},[264,423,297],{"class":270},[264,425,324],{"class":274},[264,427,428,430,432,434],{"class":266,"line":341},[264,429,330],{"class":303},[264,431,297],{"class":270},[264,433,335],{"class":309},[264,435,338],{"class":274},[264,437,438,440,442],{"class":266,"line":354},[264,439,344],{"class":303},[264,441,297],{"class":270},[264,443,444],{"class":274},"[\n",[264,446,447,450],{"class":266,"line":359},[264,448,449],{"class":309},"        \"根据这张截图和语音转写，判断用户遇到的具体问题并给排查步骤\"",[264,451,338],{"class":274},[264,453,455,458,461,464,467,469,472],{"class":266,"line":454},9,[264,456,457],{"class":274},"        types.Part.from_uri(",[264,459,460],{"class":309},"\"gs:\u002F\u002Fyour-bucket\u002Fscreenshot.png\"",[264,462,463],{"class":274},", ",[264,465,466],{"class":303},"mime_type",[264,468,297],{"class":270},[264,470,471],{"class":309},"\"image\u002Fpng\"",[264,473,474],{"class":274},"),\n",[264,476,478,480,483,485,487,489,492],{"class":266,"line":477},10,[264,479,457],{"class":274},[264,481,482],{"class":309},"\"gs:\u002F\u002Fyour-bucket\u002Faudio.wav\"",[264,484,463],{"class":274},[264,486,466],{"class":303},[264,488,297],{"class":270},[264,490,491],{"class":309},"\"audio\u002Fwav\"",[264,493,474],{"class":274},[264,495,497],{"class":266,"line":496},11,[264,498,499],{"class":274},"    ],\n",[264,501,503],{"class":266,"line":502},12,[264,504,313],{"class":274},[264,506,508,510],{"class":266,"line":507},13,[264,509,363],{"class":362},[264,511,366],{"class":274},[83,513,514],{"id":514},"流式输出",[255,516,518],{"className":257,"code":517,"language":259,"meta":260,"style":260},"for chunk in client.models.generate_content_stream(\n    model=\"gemini-3.6-flash\",\n    contents=\"用三句话总结这篇文档\",\n):\n    print(chunk.text, end=\"\")\n",[147,519,520,534,544,555,560],{"__ignoreMap":260},[264,521,522,525,528,531],{"class":266,"line":267},[264,523,524],{"class":270},"for",[264,526,527],{"class":274}," chunk ",[264,529,530],{"class":270},"in",[264,532,533],{"class":274}," client.models.generate_content_stream(\n",[264,535,536,538,540,542],{"class":266,"line":284},[264,537,330],{"class":303},[264,539,297],{"class":270},[264,541,335],{"class":309},[264,543,338],{"class":274},[264,545,546,548,550,553],{"class":266,"line":291},[264,547,344],{"class":303},[264,549,297],{"class":270},[264,551,552],{"class":309},"\"用三句话总结这篇文档\"",[264,554,338],{"class":274},[264,556,557],{"class":266,"line":316},[264,558,559],{"class":274},"):\n",[264,561,562,565,568,571,573,576],{"class":266,"line":327},[264,563,564],{"class":362},"    print",[264,566,567],{"class":274},"(chunk.text, ",[264,569,570],{"class":303},"end",[264,572,297],{"class":270},[264,574,575],{"class":309},"\"\"",[264,577,313],{"class":274},[20,579,580],{"id":580},"定价与成本核算",[582,583,584,603],"table",{},[585,586,587],"thead",{},[588,589,590,594,597,600],"tr",{},[591,592,593],"th",{},"档位",[591,595,596],{},"输入",[591,598,599],{},"输出",[591,601,602],{},"缓存输入",[604,605,606,620],"tbody",{},[588,607,608,611,614,617],{},[609,610,11],"td",{},[609,612,613],{},"$1.50 \u002F 1M",[609,615,616],{},"$7.50 \u002F 1M",[609,618,619],{},"$0.15 \u002F 1M",[588,621,622,624,627,630],{},[609,623,71],{},[609,625,626],{},"$0.30 \u002F 1M",[609,628,629],{},"$2.50 \u002F 1M",[609,631,632],{},"$0.03 \u002F 1M",[24,634,635,638],{},[28,636,637],{},"算一笔账","：假设每天 100 万次请求，每次平均 2K 输入 \u002F 500 输出 token，且前缀可缓存：",[52,640,641,644,647],{},[55,642,643],{},"输入：$1.50 × 2 ≈ $3.0（若命中缓存则降至 $0.15 × 2 ≈ $0.3）",[55,645,646],{},"输出：$7.50 × 0.5 ≈ $3.75",[55,648,649,652],{},[28,650,651],{},"单日成本约 $6.75","（命中缓存则约 $4.05），同等体量用 Pro 会贵一个数量级。",[24,654,655],{},"Flash 适合日常高并发，Flash-Lite 适合极致成本的超大规模自动化。",[20,657,658],{"id":658},"与同档模型怎么选",[582,660,661,676],{},[585,662,663],{},[588,664,665,668,670,673],{},[591,666,667],{},"维度",[591,669,11],{},[591,671,672],{},"GPT-5.6 Terra",[591,674,675],{},"Claude Sonnet 5",[604,677,678,692,705,719,733,746],{},[588,679,680,683,686,689],{},[609,681,682],{},"多模态",[609,684,685],{},"★★★★★ 原生全模态",[609,687,688],{},"reasoning + vision",[609,690,691],{},"图片",[588,693,694,697,700,703],{},[609,695,696],{},"上下文",[609,698,699],{},"1M",[609,701,702],{},"1.1M",[609,704,699],{},[588,706,707,710,713,716],{},[609,708,709],{},"输入价(\u002F1M)",[609,711,712],{},"$1.50",[609,714,715],{},"$2.50",[609,717,718],{},"$2（促销）",[588,720,721,724,727,730],{},[609,722,723],{},"编程基准",[609,725,726],{},"未单独公布",[609,728,729],{},"Terminal-Bench 84.3%",[609,731,732],{},"SWE-bench 85.2%",[588,734,735,738,741,744],{},[609,736,737],{},"延迟",[609,739,740],{},"低",[609,742,743],{},"中",[609,745,743],{},[588,747,748,751,754,757],{},[609,749,750],{},"国内可用",[609,752,753],{},"需合规通道",[609,755,756],{},"需中转",[609,758,756],{},[24,760,761,764],{},[28,762,763],{},"决策树","：",[52,766,767,774,789],{},[55,768,769,770,773],{},"要",[28,771,772],{},"多模态感知 + 长上下文 + 低成本高并发"," → Flash；",[55,775,769,776,779,780,784,785,788],{},[28,777,778],{},"硬核编程 Agent \u002F 终端自动化"," → 优先 ",[230,781,783],{"href":782},"\u002Fmodels\u002Fgpt-5-6.html","GPT-5.6"," 或 ",[230,786,675],{"href":787},"\u002Fmodels\u002Fclaude-sonnet-5.html","；",[55,790,769,791,794,795,799],{},[28,792,793],{},"极致便宜的超大规模跑量"," → Flash-Lite，或对照 ",[230,796,798],{"href":797},"\u002Fmodels\u002Fdeepseek-v4.html","DeepSeek V4","。",[20,801,802],{"id":802},"集成与工程实践建议",[52,804,805,811,823,829],{},[55,806,807,810],{},[28,808,809],{},"用输入缓存压成本","：把系统提示、知识库前缀、代码仓库上下文设为可缓存前缀，复用即享 $0.15。",[55,812,813,816,817,819,820,822],{},[28,814,815],{},"路由分层","：把简单任务留在 Flash，复杂子任务升级到 Pro\u002F",[230,818,783],{"href":782},"；这正是 ",[230,821,233],{"href":232}," 的思路。",[55,824,825,828],{},[28,826,827],{},"注意输出上限 65K","：超长生成（如整本翻译）需分段拼接。",[55,830,831,834],{},[28,832,833],{},"可观测","：记录每条请求的模态组合与 token 分布，多模态请求成本波动大，建议单独打点。",[20,836,837],{"id":837},"避坑清单",[52,839,840,846,852,858],{},[55,841,842,845],{},[28,843,844],{},"别把它当编程主力","：发布未给 SWE-bench \u002F Terminal-Bench 等分数，强代码 Agent 场景先用 GPT-5.6 \u002F Claude 验证。",[55,847,848,851],{},[28,849,850],{},"缓存不是默认生效","：必须显式构造可复用前缀，否则一直走 $1.50 全价。",[55,853,854,857],{},[28,855,856],{},"1M 上下文不是免费午餐","：单次长上下文调用成本随长度线性上升，按需截取。",[55,859,860,863],{},[28,861,862],{},"国内接入","：走 Google AI Studio \u002F Vertex AI，需合规通道；别假设能直连。",[20,865,867],{"id":866},"faq","FAQ",[24,869,870,873],{},[28,871,872],{},"Q: Flash 和 Pro 怎么选？","\nA: 90% 的生产场景 Flash 够用且便宜一个量级。只在「质量天花板」卡脖子（复杂推理、最难的多模态理解）时才上 Pro。",[24,875,876,879],{},[28,877,878],{},"Q: 65K 输出够吗？","\nA: 对摘要、对话、抽取类任务绰绰有余；整本文档生成、超长代码重构需要分段。",[24,881,882,885],{},[28,883,884],{},"Q: 国内怎么接入？","\nA: 通过 Vertex AI 或合规中转；官方 AI Studio 需合规网络环境。",[24,887,888,891],{},[28,889,890],{},"Q: 能替代编程 Agent 吗？","\nA: 不适合当主力。多模态与长上下文很强，但编程基准未公布，硬核编码先对照 GPT-5.6 \u002F Claude 的公开成绩。",[20,893,894],{"id":894},"延伸阅读",[52,896,897,911,923],{},[55,898,899,900,902,903,902,905,902,909],{},"同档对照：",[230,901,783],{"href":782}," \u002F ",[230,904,675],{"href":787},[230,906,908],{"href":907},"\u002Fmodels\u002Fgemini-3-pro.html","Gemini 3 Pro",[230,910,798],{"href":797},[55,912,913,914,902,917,902,919],{},"配套工具：",[230,915,916],{"href":232},"Cursor",[230,918,245],{"href":244},[230,920,922],{"href":921},"\u002Fcoding\u002Fcli\u002Fclaude-code.html","Claude Code",[55,924,925,926],{},"相关资讯：",[230,927,929],{"href":928},"\u002Fnews\u002F2026\u002Fgemini-3-6-flash-release.html","Gemini 3.6 Flash 发布",[931,932,933],"style",{},"html pre.shiki code .szBVR, html code.shiki .szBVR{--shiki-default:#D73A49;--shiki-dark:#F97583}html pre.shiki code .sVt8B, html code.shiki .sVt8B{--shiki-default:#24292E;--shiki-dark:#E1E4E8}html pre.shiki code .s4XuR, html code.shiki .s4XuR{--shiki-default:#E36209;--shiki-dark:#FFAB70}html pre.shiki code .sZZnC, html code.shiki .sZZnC{--shiki-default:#032F62;--shiki-dark:#9ECBFF}html pre.shiki code .sj4cs, html code.shiki .sj4cs{--shiki-default:#005CC5;--shiki-dark:#79B8FF}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}html.dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}",{"title":260,"searchDepth":291,"depth":291,"links":935},[936,937,943,949,954,955,956,957,958,959],{"id":22,"depth":284,"text":22},{"id":81,"depth":284,"text":81,"children":938},[939,940,941,942],{"id":85,"depth":291,"text":86},{"id":118,"depth":291,"text":118},{"id":153,"depth":291,"text":153},{"id":178,"depth":291,"text":178},{"id":207,"depth":284,"text":207,"children":944},[945,946,947,948],{"id":210,"depth":291,"text":211},{"id":217,"depth":291,"text":218},{"id":224,"depth":291,"text":225},{"id":237,"depth":291,"text":238},{"id":249,"depth":284,"text":250,"children":950},[951,952,953],{"id":253,"depth":291,"text":253},{"id":369,"depth":291,"text":370},{"id":514,"depth":291,"text":514},{"id":580,"depth":284,"text":580},{"id":658,"depth":284,"text":658},{"id":802,"depth":284,"text":802},{"id":837,"depth":284,"text":837},{"id":866,"depth":284,"text":867},{"id":894,"depth":284,"text":894},"multimodal",1048576,"Google 于 2026 年 7 月 21 日发布的稳定版 Flash 模型，1M 上下文、65K 输出，定价 $1.50\u002F$7.50 每百万 token，定位高吞吐多模态场景，原生支持函数调用与流式。","md",65536,{},"\u002Fmodels\u002Fgemini-3-6-flash","$1.50 \u002F $7.50（cache $0.15）\u002F 百万 token","2026-07-27",[970,971],"coding\u002Fide\u002Fcursor","coding\u002Fapi\u002Fopenrouter","2026-07-21",{"title":11,"description":962},"gemini-3-6-flash","models\u002Fgemini-3-6-flash",[977,978,979,980,981],"原生多模态：文本 \u002F 图像 \u002F 视频 \u002F 音频统一输入，响应快","1M 输入上下文 + 65K 输出，长文档与代码库友好","Flash 定位低价低延迟，适合高吞吐自动化与 Agent","Google AI Studio \u002F Vertex AI 全托管，接入简单","原生函数调用、结构化输出与流式，工程接入成本低",[983,984,985,986],"高并发多模态内容生成与审核","长上下文 RAG \u002F 文档理解","低成本 Agent 与自动化流水线","实时翻译、摘要、客服对话","Google",[989,990,991],"独立于编程（SWE-bench \u002F Terminal-Bench 等）基准未随发布公布，编程 Agent 场景仍以 GPT-5.6 \u002F Claude 为主流参照","输入缓存需命中才享 $0.15，短请求收益有限","国内直连需合规通道（AI Studio \u002F Vertex AI）","aG_eV1XiNF7ywOhloYfXbTDiDLwB6ZofUcfcltDko3Y",1785161585337]