2026年6月10日，Google I/O大会在山景城的 Shoreline Amphitheatre 拉开帷幕。与往年不同，这次的重头戏不是Android的更新或Pixel新机的亮相，而是Google对AI赛道发起的全面进攻。CEO桑达尔·皮查伊（Sundar Pichai）站在舞台中央，向全球开发者宣布了两款全新的Gemini系列模型——Gemini 3.5 Flash和全模态Gemini Omni。
On June 10, 2026, Google I/O opened at Shoreline Amphitheatre in Mountain View. Unlike previous years, the main attraction this time wasn't Android updates or new Pixel devices, but Google's full-scale offensive in the AI space. CEO Sundar Pichai stood center stage and announced two brand-new models in the Gemini series to global developers: Gemini 3.5 Flash and the fully multimodal Gemini Omni.
Gemini 3.5 Flash：速度即正义
Gemini 3.5 Flash: Speed Is Justice
Gemini 3.5 Flash是Google在生成速度上追求极致的一款模型。根据Google AI实验室公布的基准测试数据，在编码任务中，Gemini 3.5 Flash的首token响应时间比Gemini 2.5 Flash缩短了72%，整体代码生成速度提升了4倍。在人类评价盲测中，Gemini 3.5 Flash生成的代码在功能正确率和代码质量方面与GPT-4o的表现不相上下，但在推理耗时上快了约60%。
Gemini 3.5 Flash is a model designed by Google for extreme speed in generation. According to benchmark data published by Google AI Lab, in coding tasks, Gemini 3.5 Flash reduced first-token response time by 72% compared to Gemini 2.5 Flash, with overall code generation speed improving 4x. In human evaluation blind tests, the code generated by Gemini 3.5 Flash matched GPT-4o in functional correctness and code quality, but took approximately 60% less inference time.
Google云部门高级副总裁Thomas Kurian在发布会上表示："速度不是次要指标，而是AI落地的核心门槛。一个再聪明的模型，如果响应需要等待十秒以上，开发者就不会用它。"为此，Google宣布将Gemini 3.5 Flash API以免费额度大幅提高的方式向所有开发者开放，同时将推理成本降低至Gemini 2.5的三分之一。
Thomas Kurian, Senior VP of Google Cloud, stated at the event: "Speed isn't a secondary metric; it's the core barrier to AI adoption. No matter how smart a model is, if responses take more than ten seconds, developers won't use it." To this end, Google announced it would significantly increase the free tier for the Gemini 3.5 Flash API to all developers and reduce inference costs to one-third of Gemini 2.5.
Gemini Omni：从"多模态"到"全模态"的跨越
Gemini Omni: The Leap from "Multimodal" to "Fully Multimodal"
如果说Gemini 3.5 Flash是Google的"速度武器"，那么全模态Gemini Omni就是其"能力核弹"。Google首次将文本、图像、音频、视频、3D和传感器数据统一到一个模型架构中，实现真正的端到端全模态理解与生成。这意味着Gemini Omni可以同时"看"一段视频、"听"其中的语音、"读"屏幕上的文字，并综合所有信息进行推理和回应。
If Gemini 3.5 Flash is Google's "speed weapon," then the fully multimodal Gemini Omni is its "capability nuclear bomb." Google unified text, images, audio, video, 3D, and sensor data into a single model architecture for the first time, achieving true end-to-end fully multimodal understanding and generation. This means Gemini Omni can simultaneously "watch" a video, "hear" its audio, "read" on-screen text, and integrate all information for reasoning and response.
Google在发布会上演示了一个医疗场景：Gemini Omni可以同时分析患者的CT影像、病历文本、语音问诊录音和心率传感器数据，在不到5秒内生成一份综合诊断建议。Google深时技术（Google DeepMind）首席研究员Demis Hassabis表示："这不是将多个模型拼接在一起，而是从架构层面实现全模态的统一。这将是AI历史上的一个里程碑。"
Google demonstrated a medical scenario at the event: Gemini Omni can simultaneously analyze a patient's CT images, medical records, voice consultation recordings, and heart rate sensor data, generating a comprehensive diagnostic recommendation in under 5 seconds. Demis Hassabis, Chief Researcher at Google DeepMind, stated: "This isn't stitching multiple models together, but achieving full multimodal unity at the architecture level. This will be a milestone in AI history."
正面硬刚：Google的AI"三路出击"战略
Direct Confrontation: Google's AI "Three-Pronged" Strategy
这次I/O大会清晰地展示了Google在AI赛道的"三路出击"战略：以Gemini 3.5 Flash在速度和成本上抢占开发者市场，以Gemini Omni在能力上对标GPT-5和Claude 4，同时配合Google Workspace的深度AI整合（新版Gemini在Docs、Sheets、Gmail中的原生嵌入），在应用层面形成与Microsoft Copilot生态的正面竞争。
This I/O conference clearly demonstrated Google's "three-pronged" AI strategy: using Gemini 3.5 Flash to capture the developer market on speed and cost, Gemini Omni to benchmark against GPT-5 and Claude 4 on capability, while simultaneously deepening AI integration in Google Workspace (native embedding of the new Gemini in Docs, Sheets, Gmail), forming direct competition with the Microsoft Copilot ecosystem at the application level.
一位不愿透露姓名的AI行业分析师告诉《科技周刊》："Google这次在I/O上投入的展示内容，比过去两届加起来还要多。它传递的信息非常直接：Google不是AI竞赛的追随者，Google要重新定义规则。"
An AI industry analyst, speaking on condition of anonymity to Tech Weekly, stated: "Google's presentation content at this I/O exceeds the previous two editions combined. The message is very direct: Google isn't a follower in the AI race; Google wants to redefine the rules."
开发者反应：拥抱还是观望？
Developer Response: Embrace or Observe?
I/O大会结束后，开发者社区的反应总体积极。在GitHub上，以Gemini为前缀的仓库在过去24小时内新增了超过2,300个，数量超过了GPT和Claude相关仓库的总和。AI编程工具Vercel的CTO Will Vincent在X平台上发文称："Gemini 3.5 Flash的API响应速度令人震惊，我们已经将其集成到下一个版本的Vercel AI SDK中。"
After the I/O conference, developer community reactions were overall positive. On GitHub, repositories prefixed with "Gemini" added over 2,300 new entries in the past 24 hours, exceeding the combined total of GPT and Claude related repositories. Vercel CTO Will Vincent posted on X: "Gemini 3.5 Flash's API response speed is astonishing; we've already integrated it into the next version of Vercel AI SDK."
不过也有部分开发者保持谨慎。独立AI研究员Karpathi指出："Google模型的速度确实很快，但在复杂推理和创造性任务上的表现，还需要更多独立第三方评测来验证。我们不应该因为一个惊艳的发布会Demo就过早下结论。"这一观点得到了不少人的认同——在AI行业，"Demo Day"与"Product Day"之间始终隔着一道巨大的鸿沟。
However, some developers remain cautious. Independent AI researcher Karpathi pointed out: "Google's models are indeed fast, but performance in complex reasoning and creative tasks still needs more independent third-party evaluations to verify. We shouldn't jump to conclusions too early based on an impressive keynote demo." This view resonated with many -- in the AI industry, there's always a vast gulf between "Demo Day" and "Product Day."