[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-gemini-3-8-audio-live-extended-thinking":3,"topics-all":38,"news-related-8303f420-b3f1-485a-aa6f-7775256c84a7":57},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"8303f420-b3f1-485a-aa6f-7775256c84a7","Gemini 3.8 Audio 双发:Live 和 Extended Thinking 把思考+说话压到近实时","Google DeepMind 发布 Gemini 3.8 Audio 双模型 Live 与 Live Extended Thinking,基于 Gemini 3 Pro 原生输出音频,把思考过程实时说出来,部署到 API\u002FAI Studio\u002FVertex\u002FWorkspace 全通道。","2026 年 9 月 15 日,Google DeepMind 在 model-cards 频道放出两个新模型 Gemini 3.8 Live 和 Gemini 3.8 Live Extended Thinking,挂在同一张 \"Gemini 3.8 Audio\" 总卡下。这次的看点不是参数,也不是基准跑分 —— 是 Google 把\"实时对话 + 边想边说\"这两条过去分裂的轨道,合并到同一条音频输出流里。\n\n## 不是什么新架构,是 Gemini 3 Pro 的音频能力解锁\n\n先把架构讲清楚:这两个模型不是从零训出来的。Model Card 直接写明,「Gemini 3.8 Audio is based on Gemini 3 Pro」,从训练数据、训练处理、硬件到软件实现,整张卡都把读者指向 Gemini 3 Pro 的对应 model card。所谓\"Live\"和\"Live Extended Thinking\"本质是 Gemini 3 Pro 在\"原生多模态\"基础上,把\"音频\"从支持模态升级成\"原生输出模态\" —— 模型不再依赖外接 TTS 把文字读出来,而是直接端到端生成语音 token。\n\n输入端,两个模型吃 128K token 上下文,模态包括音频、图像、视频和文本,音频是 16-bit PCM、16 kHz 的原始流;输出端,文本 + 音频一起出,文本 token 上限 64K,音频是 16-bit PCM、24 kHz。架构上模型走 40 层 Transformer,但模型卡明确说\"和 Gemini 3 Pro 共用 backbone 细节\",所以这里没有新论文级别的结构创新,真正的工程量在\"流式多模态 token 调度\"和\"边生成边说\"的延迟优化上。\n\n## 双档发布:一个省成本,一个能\"想一会儿再说\"\n\n两个模型虽然同源,但定位明显分层。DeepMind 官方音频页把它们的差异写得很直白:\n\n- **Gemini 3.8 Live**:面向\"高吞吐、成本敏感、近实时语音接口\",典型场景是客服机器人、电商导购、虚拟 NPC 实时陪聊这种\"对话本身不需要多复杂推理,但必须秒回\"的场景。\n- **Gemini 3.8 Live Extended Thinking**:面向\"复杂推理 + 多步任务\",主打\"边思考边把思考过程说出来\"。模型在生成语音回答之前,会有一段\"thinking\"间隔,然后一边想一边用自然语气的语音把推理进度\"念\"出来 —— 而不是像传统链式推理那样只在文本侧 think,最后用 TTS 念结果。\n\n第二个档位是这次最有意思的产品决策。当下绝大多数\"会思考\"的 AI 助手(无论是 OpenAI 的 o 系列还是 Anthropic 的 thinking 模式)都有一个明显的体验断点:模型先静默想几十秒,突然开始说话;用户体验上像\"等了半天突然开口\",中间那段等待是\"模型在想\"还是\"模型卡了\"完全没反馈。Extended Thinking 把这段\"思考期\"做成\"可听的思考\" —— 用户能听到模型在自言自语,实时观察它在哪个方向上走。这种\"过程可见的推理\"在 OpenAI 和 Anthropic 的当前主力产品里都还没做到音频侧,Google 这次是把 thinking 模式直接拉到了语音流里。\n\n## 部署通道:从 Gemini App 到 Gmail 都接上\n\n模型卡列出的分发渠道基本是 Google 全家桶:\n\n- 3.8 Live: Gemini API、Gemini App、Google AI Studio、Google Cloud \u002F Vertex AI、Google Search Live\n- 3.8 Live Extended Thinking: 上面四个 + Google Workspace(Gmail、Docs、Keep)\n\nWorkspace 这一档额外集成是真正落地层面的信号 —— Extended Thinking 不只是给开发者做 agent 用的,Gmail \u002F Docs \u002F Keep 三个消费级 Workspace 应用会直接拿到这个\"边想边说\"能力。这意味着 Google 准备把\"会思考的语音助手\"塞进用户每天打开 20 次的邮件和文档工作流,而不是停留在 AI Studio demo 阶段。\n\n研发侧的接入门槛也被刻意压低。Live API 文档页(ai.google.dev\u002Fgemini-api\u002Fdocs\u002Flive-api,最近更新 2026\u002F09\u002F10)把\"WebSocket 双向流\"列为默认协议,GenAI SDK、WebSocket 直连、Agent Development Kit(ADK)Streaming 三种实现路径都给完整示例代码。第三方集成方面,LiveKit、Pipecat、Fishjam、Voximplant、Agora、Firebase AI Logic 全部在合作伙伴名单里 —— 对实时语音 agent 开发者来说,这个分发密度和 Google 自家搜索、AI Studio 拉通,意味着\"接 Gemini 3.8 Live\"不是另起炉灶,而是在现有 WebRTC \u002F WebSocket 实时栈上换一行 model id。\n\n## 评测与安全:Sierra τ³-Bench 进官方方法学\n\nModel card 在评估这块很克制,没有贴一长串基准表,而是引向专门的 Evals & Methodology 子页(deepmind.google\u002Fmodels\u002Fevals-methodology\u002Fgemini-3-8-live),实际评测由三家独立机构跑:\n\n- **ServiceNow EVA-Bench**:多轮 bot-to-bot 语音对话框架,用 Gemini Enterprise Agent Platform 跑,跑 thinking-minimal 和 thinking-high 两档。\n- **Artificial Analysis**:第三方独立评测,跑的是 speech-to-speech 维度,看推理质量、对话动力学、生成时延、价格的综合表现。\n- **Sierra τ³-Bench**:测模型在大型非结构化知识库上做多步 tool call、解决真实银行工作流的能力,跑 Gemini API,model-id 分别对应 gemini-3.8-live-preview 和 gemini-3.8-live-extended-thinking,thinking-high \u002F 默认采样。\n\n安全方面,所有音频输出自动挂 SynthID 水印(Google 自家音频溯源技术),前沿安全评估沿用 Gemini 3.7 Flash 的结论,认为 3.8 Live \u002F Extended Thinking \"相比 3.7 Flash 没有实质性能力跃升\",因此不触发 Tracked \u002F Critical Capability Levels。知识截止 2025 年 1 月。\n\n## 行业意义:把\"语音流里的推理\"做成产品力\n\n把这件事放到更大的行业语境里看:2026 年下半年,所有前沿模型厂商的差异化竞争已经从\"谁更长\"转向\"谁更能在多模态流里稳住\"。GPT-Realtime 2.1、Gemini 3.1 Flash TTS、Inworld TTS-2、Cartesia Sonic 3.6 这一连串实时音频\u002F语音模型在 5-9 月密集发布,说明实时语音这条赛道已经从\"能不能做\"过渡到\"怎么做才能不让用户等\"。\n\nGemini 3.8 Live Extended Thinking 在这个赛道里走了不同方向 —— 不卷延迟数字,卷\"思考过程对用户可见\"。这种\"出声的 chain-of-thought\"短期看会拉高单次响应时间(用户得听模型在自言自语),但长期看是把\"模型在做什么\"这层可解释性,从开发者的 log 里搬到了终端用户的耳朵里。如果 Extended Thinking 这个范式被验证有用,OpenAI 和 Anthropic 跟进几乎是确定事件 —— 届时\"会说话的 thinking 模型\"可能会变成下一代语音助手的默认形态。\n\n对国内厂商的参照:目前国产实时语音模型(阿里 CosyVoice、字节 Seed-TTS、腾讯混元 TTS 等)主要卷\"声音质量 + 实时率\",还没有把 thinking 模式端到端拉到语音流里。如果 2026 年 Q4 之前有国产模型把\"边想边说\"做出来,产品上能直接吃 Google 这套范式的红利。\n\n## 所以呢\n\nGemini 3.8 Audio 的发布,关键不在模型本身 —— 它就是 Gemini 3 Pro 加了原生音频输出。关键在于\"原生输出 + thinking 可见 + Workspace 全家桶集成\"这三件事合起来,Google 把\"实时语音 + 推理可解释性 + 消费级落地通道\"第一次打包给到用户。开发者可以直接调 API,普通用户未来几个月会在 Gmail、Docs 里听到\"模型在想什么\"。这是实时语音 agent 赛道从 demo 走向日常工具的标志性一步。","https:\u002F\u002Fdeepmind.google\u002Fmodels\u002Fmodel-cards\u002Fgemini-3-8-audio\u002F","35ce748f-48b7-4638-88ef-effa57a7e749",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"a9524a82-a7c5-4daa-bb4b-a7ee77bb0b94","gemini",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"8cf7490f-2449-4ba7-be19-61befa0d92b4","google",{"id":19,"name":20,"slug":20,"description":14,"color":14},"7e89b5cc-57db-4f37-bc6d-28919a73931c","model-release",{"id":22,"name":23,"slug":23,"description":14,"color":14},"499f4b56-819d-49a3-9609-33e775143b86","multimodal",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"d12b8894-c459-4c14-b103-fe4d9115338b","en","Gemini 3.8 Audio ships: Live and Extended Thinking fold reasoning into near-real-time speech","Google DeepMind ships Gemini 3.8 Live and Live Extended Thinking, two audio-native variants of Gemini 3 Pro. The Extended Thinking tier speaks its reasoning out loud in near real time, with distribution to API, AI Studio, Vertex, Search Live and Workspace.","On September 15, 2026, Google DeepMind added two new models — Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking — to its model-cards channel, both published under a single \"Gemini 3.8 Audio\" umbrella card. The story is not new parameters and not new benchmark numbers. The story is that Google has merged two previously separate tracks — real-time conversation and audible reasoning — into a single audio output stream.\n\n## Not a new architecture, just Gemini 3 Pro with native audio unlocked\n\nLet me get the architecture out of the way first. Neither model is trained from scratch. The model card states plainly that \"Gemini 3.8 Audio is based on Gemini 3 Pro,\" and points readers at the Gemini 3 Pro card for training data, training processing, hardware and software details. The \"Live\" and \"Live Extended Thinking\" labels essentially mean that Google has promoted \"audio\" from a supported input modality to a native output modality on top of Gemini 3 Pro's native multimodal stack. The model no longer hands text to an external TTS — it generates speech tokens end-to-end.\n\nOn the input side, both models accept 128K-token context, with modalities covering audio, image, video and text (16-bit PCM audio at 16 kHz). On the output side, the model emits text plus audio together, with 64K tokens of text headroom and 16-bit PCM audio at 24 kHz. Architecturally the model is a 40-layer Transformer, but the model card explicitly defers backbone details to the Gemini 3 Pro card. There is no paper-level structural innovation here — the real engineering work is in streaming multimodal token scheduling and in the latency optimization that makes \"generate while speaking\" actually feel real-time.\n\n## Two tiers: cheap and fast, or think-out-loud\n\nSame backbone, but the two models are clearly split. DeepMind's official audio page states the distinction bluntly:\n\n- **Gemini 3.8 Live** is positioned for \"high-throughput, cost-sensitive, near real-time voice interfaces\" — the typical scenario is customer-service bots, e-commerce shopping assistants, NPC companions in games, anything where the conversation itself does not need complex reasoning but must respond in under a second.\n- **Gemini 3.8 Live Extended Thinking** is positioned for \"complex reasoning and high-complexity tasks\" — its signature move is to \"think out loud\": before it starts speaking, the model enters a thinking interval, and then narrates its own reasoning progress in real time with a natural voice, instead of thinking silently and only at the very end pushing a TTS-read summary.\n\nThe second tier is the most interesting product decision here. Almost every \"thinking\" AI assistant today — whether OpenAI's o-series or Anthropic's thinking mode — has the same experience break: the model sits silent for tens of seconds, then suddenly starts talking. From the user's perspective, there is no feedback signal that says \"the model is thinking\" versus \"the model is stuck.\" Extended Thinking turns that silent thinking interval into audible thinking. The user can hear the model mumble to itself, watching it change direction in real time. This \"process-visible reasoning\" has not been done on the audio side in OpenAI's or Anthropic's current flagship products. Google is the first to drag thinking mode into the speech stream itself.\n\n## Distribution: from Gemini App to Gmail and Docs\n\nThe model card lists Google-as-distributor across the board:\n\n- For 3.8 Live: Gemini API, Gemini App, Google AI Studio, Google Cloud \u002F Vertex AI, and Google Search Live.\n- For 3.8 Live Extended Thinking: the same four plus Google Workspace (Gmail, Docs, Keep).\n\nThat extra Workspace integration is the real landing signal — Extended Thinking is not just for developers building agents. Gmail, Docs and Keep are the three Workspace apps users open twenty times a day. Google is preparing to push \"thinking-out-loud voice assistants\" into the daily email and document workflow, not just AI Studio demos.\n\nThe integration bar for developers has been deliberately lowered. The Live API docs page (ai.google.dev\u002Fgemini-api\u002Fdocs\u002Flive-api, last updated 2026-09-10) makes WebSocket duplex streaming the default protocol and ships full sample code for the GenAI SDK, raw WebSocket, and the Agent Development Kit (ADK) Streaming path. On the third-party side, LiveKit, Pipecat, Fishjam, Voximplant, Agora, and Firebase AI Logic are all in the partner list. For real-time voice-agent developers, this distribution density plus Google's own Search and AI Studio surface means switching to Gemini 3.8 Live is not a rewrite — it is a one-line model-id swap on top of an existing WebRTC \u002F WebSocket real-time stack.\n\n## Evaluation and safety: Sierra τ³-Bench makes it into the official methodology\n\nThe model card stays restrained on evaluation, and points to a dedicated Evals & Methodology subpage (deepmind.google\u002Fmodels\u002Fevals-methodology\u002Fgemini-3-8-live). The actual benchmarks are run by three independent parties:\n\n- **ServiceNow EVA-Bench** — a multi-turn bot-to-bot spoken conversation framework, run on the Gemini Enterprise Agent Platform with both thinking-minimal and thinking-high configurations.\n- **Artificial Analysis** — an independent third-party benchmark that measures speech-to-speech models on reasoning quality, conversational dynamics, generation latency, and price.\n- **Sierra τ³-Bench** — tests whether agents can navigate a large unstructured knowledge base and execute multi-step tool calls to resolve realistic banking workflows. The runs use the Gemini API with model-ids `gemini-3.8-live-preview` and `gemini-3.8-live-extended-thinking`, under thinking-high and default sampling.\n\nOn safety, every audio output is automatically tagged with SynthID, Google's own audio watermarking technology. The Frontier Safety assessment carries forward Gemini 3.7 Flash's conclusion: 3.8 Live and Extended Thinking do not represent a material capability jump over 3.7 Flash, so neither model triggers any Tracked or Critical Capability Level. The knowledge cutoff is January 2025.\n\n## Industry meaning: \"reasoning in the audio stream\" as a product feature\n\nZoom out: from the second half of 2026 onward, every frontier model lab has moved differentiation away from \"who is longer\" and toward \"who can hold steady inside a multimodal stream.\" GPT-Realtime 2.1, Gemini 3.1 Flash TTS, Inworld TTS-2, Cartesia Sonic 3.6 — this parade of real-time audio and speech models shipped in May through September, and it tells you the real-time voice track has already moved past \"can it be done\" and is now squarely in the \"how do you make it stop making the user wait\" phase.\n\nGemini 3.8 Live Extended Thinking takes a different direction in that race — it does not chase latency numbers, it chases \"visible reasoning.\" Out-loud chain-of-thought will increase the per-turn response time (the user has to listen to the model talk to itself), but over the longer arc it relocates \"what is the model doing\" from developer logs into the end-user's ear. If this paradigm is validated, OpenAI and Anthropic following suit is almost certain. When that happens, \"thinking-out-loud speech models\" will likely become the default form of the next generation of voice assistants.\n\nFor Chinese vendors, the comparison is direct. Today the domestic real-time speech models — Alibaba CosyVoice, ByteDance Seed-TTS, Tencent Hunyuan TTS, and so on — are mostly optimizing on \"voice quality + real-time ratio.\" None of them have pulled thinking mode end-to-end into the speech stream. If any of them ships \"think-while-speak\" before the end of Q4 2026, they will get to eat the product-level dividend that this Google pattern sets up.\n\n## So what\n\nThe Gemini 3.8 Audio release is not about a new model — it is Gemini 3 Pro with a native audio output. The substance is the combination of three things: native audio output, visible thinking, and a Workspace-wide distribution channel. For the first time, Google packages \"real-time voice + reasoning interpretability + consumer-grade distribution\" as a single product surface. Developers can hit the API directly, and ordinary users will soon hear what their model is actually thinking inside Gmail and Docs. This is the marker that the real-time voice-agent track is moving from demo to daily tool.","gemini-3-8-audio-live-extended-thinking","2026-09-16T03:00:00Z","2026-09-16T03:07:06.411523Z","2026-09-16T03:07:06.411536Z",true,"agent",76,[39,48],{"slug":40,"tag_slug":40,"title_zh":41,"title_en":42,"intro_zh":43,"intro_en":44,"id":45,"is_active":35,"created_at":46,"modified_at":47},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":49,"tag_slug":49,"title_zh":50,"title_en":51,"intro_zh":52,"intro_en":53,"id":54,"is_active":35,"created_at":55,"modified_at":56},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":58},[59,64,69,74,79,84],{"id":60,"title":61,"news_slug":62,"published_at":63},"039ff515-68e7-4f11-866a-1da97e26eb45","Gemini 3.8 Live 拿下 S2S 实时语音榜第一","gemini-3-8-live-voice-s2s-number-one","2026-09-15T17:00:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"1ba7e499-d93d-4566-bbeb-0b762904c0ab","Google 把 Lyria 3.5 装进 Gemini 与公开 API:音乐生成从独立工具变成默认选项","lyria-3-5-gemini-app-api","2026-09-06T23:06:32+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"9af3dd83-6ed9-498d-9da0-547d917f3e19","语音转文字有了专用模型:Gemini 3.5 Transcribe 上线,出稿快 70%","gemini-35-transcribe-dedicated-asr","2026-08-31T13:30:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"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":80,"title":81,"news_slug":82,"published_at":83},"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":85,"title":86,"news_slug":87,"published_at":88},"e0a484e4-41e0-4f91-9b9b-a196bbdcf3ba","Gemini 3.8 Flash 双发:同价升级 + Cyber 走可信项目 Fairwind","gemini-3-8-flash-cyber-fairwind-launch","2026-09-03T03:00:00+00:00"]