Gemini 3.6 Flash
最大上下文
1M
最大输出
65.5K
发布日期
暂无
gemini-4-argonGemini 4 Argon 是 Google 的旗舰推理模型,专为复杂软件工程、深度数学与科学问题求解,以及自主多步智能体而打造。Google 训练它能在冗长多步任务中保持上下文,并将混合文本、表格、图表和图像的长文档转化为单一有据可依的输出。它在金融、法律、编程和税务工作中领跑 Vals Index,并在单次响应中支持最长 100 万输出 token 的长解码。
上下文窗口
1M Token
最大输出
65.5K Token
发布日期
2026年10月13日
模态
| 输入价格 | 输出价格 | 缓存读取 | 缓存创建 5m |
|---|---|---|---|
| $4/M | $20/M | $0.2/M | $0.375/M |
推理
工具调用
Temperature 参数
附件
知识库
Endpoint 协议
面向复杂推理、持续工程任务与多模态理解。
面向高难度编程、数学与科学问题,可调节推理深度。
旨在跨复杂、多步骤的软件工程与自主智能体流程持续保持上下文。
输入、推理和输出共享 1,048,576 Token;Google 的长输出流程可突破标准 64K 输出上限。
融合文本、图像、音频、视频和 PDF 输入,生成文本形式的分析与综合结果。
适用于复杂开发、研究与文档分析的实际任务。
规划并实施跨文件重构或遗留代码迁移,再依据项目要求审查修改。
将包含文字、表格、图表和图像的长文档整理为依据所提供材料撰写的连贯报告。
探索高难度技术问题、推导数学论证,并起草供独立验证的研究分析。
构建跨多个步骤完成规划、创建、检查和修订的软件或 3D 原型流程。
import OpenAI from "openai"
const client = new OpenAI({
apiKey: process.env.TOKENHUB_API_KEY,
baseURL: "https://us-api.tokenhub.com/v1",
})
const result = await client.chat.completions.create({})
console.log(result.choices[0]?.message?.content)Gemini 4 Argon (High)
| 指数分数 | ||
|---|---|---|
| Artificial Analysis 综合能力指数 | Artificial Analysis 综合能力评分 | 52.6 |
指标来源 Artificial Analysis
了解模型能力、实际取舍与 TokenHub 接入方式。
它是 Google DeepMind 面向复杂编程、自主智能体、数学与科学任务的旗舰推理模型。
适用于需要持续推理的跨文件工程、多模态文档综合和技术研究。
Google 文档列出文本、图像、音频、视频和 PDF 输入,输出为文本。图像与音频生成需要其他模型。
模型可能产生幻觉,复杂请求也可能较慢。百万 Token 预算由输入、推理和输出共享;超过 64K 的输出需要 Google 长输出流程及路由支持。
若账户中已上架,请选择 Gemini 4 Argon,使用 TokenHub 展示的模型 ID、接口地址和 API 密钥;接入前确认所选路由的功能。
Google 已提供付费 API 访问。TokenHub 的价格、路由和账户可用性请查看当前模型页,可能与 Google 直连定价不同。
当任务重视推理深度、大规模混合模态输入或持续工程流程时可选用它。请用代表性任务比较准确性、延迟和总 Token 成本。
通过一个 API Key 接入 Gemini 4 Argon 与更多 AI 模型。
Gemini 4 Argon 媒体与演示
已核验的官方帖子与演示,以及社区早期使用体验。
X (Twitter)
Gemini 4 Argon 应用上线公告
Google Gemini ·
官方公告介绍 Gemini 4 Argon 在 Gemini 应用中的推出。
The moment you've been waiting 4… Our new frontier model, Gemini 4 Argon, is rolling out now in the Gemini app to bring your most complex ideas into reality. See how Gemini 4 Argon and our latest updates across the Gemini app help you get the most out of every prompt. 🧵
查看原文Gemini 4 Argon 从规划到成果
Google Gemini ·
展示 Gemini 应用使用 Gemini 4 Argon 制作网页应用、3D 文件、幻灯片和报告的流程。
Gemini 4 Argon breaks down your tasks step-by-step and instead of stopping at a text outline, Argon helps you build the real thing, whether that means: ➡️ Generating a shareable web app ➡️ Giving you ready-to-print 3D files ➡️ Producing slides and reports that are ready to go https://t.co/V4RPXun8I9
https://pbs.twimg.com/amplify_video_thumb/2110039533853212673/img/r4OrZZO4Lsf8FhNo.jpg查看原文Gemini 4 Argon 在 Gemini 应用中编程
Google Gemini ·
介绍编程能力改进,以及 macOS 版 Gemini 应用中的 Antigravity 编程界面。
Gemini 4 Argon also brings major improvements in coding, so we’re making it easier to switch between chatting and coding with Google Antigravity's agentic coding interface now available directly in the Gemini app for macOS. https://t.co/waB3VmjmDI
https://pbs.twimg.com/amplify_video_thumb/2110039768579080193/img/q_yj8_IP7nVoDQHf.jpg查看原文Reddit
Gemini 4 Argon:编程初步评测
Gemini 4 Argon Initial review
this-is-hilarours ·
社区用户比较后端实现方案,评论讨论测试方法与不同的编程体验。
展开全文
As an ultra user just got the access to Gemini 4 Argon. i made an initial test in my python backend project . the project is already in production and there was a real feature integration that i have started working today . so i was curious how it would compare against Opus 5.5 . I have a claude code subscription and i also use a factory droid subscription . so i created an implementation plan based on the spec with opus 5.5 with extra high effort level . similarly created an implementation plan with Argon high . Both of the plan was fed to Gpt 6 astra who worked as judge as i believe gpt model always works as strong review model . i asked it to carefully judge and give both of them score out of 100 . here is the score astra give Opus 5.5 Score - 77/100 Gemini 4 Argon Scores - 59/100 I gave the original score and review to Gemini in antigravity it agreed with all of that . i think more testing would be needed from my side to test it properly . but it needs at least one more reasoning level like extra high One thing surprised me It compacted the conversation relatively faster . i don't know what is the context window set here but previously i never got the context compacted this fast . pretty sure the context is not 1 million current set in antigravity app
Gemini 4 Argon:发布讨论
Gemini 4 Argon Release Blog
elemental-mind ·
社区帖子分享 Google 关于 Gemini 4 Argon 扩大开放范围的公告。
查看原文Gemini 4 Argon:早期用户反馈
Thanks for Gemini 4 Argon
Liberation2020 ·
一位 Ultra 订阅者分享积极的任务体验,讨论涉及速度、用量限制和个人工作流程。
展开全文
I’m making this for the Google team that may be checking the thread. I’m an Ultra user and have patiently been waiting for this model. It is everything I need and more. I’ve been seeing others getting great results from it. As well. It’s ability to turn some of my things into skills that actually work is next level compared to what was happening before. It’s work rate is also impressive in my opinion detailed heavy task that I feel should maybe take a little while are done quite quickly. I’m still fine running my request style with it but I am quite happy with it this past couple of hours. It even did better on a task I had set to Astra to complete. (Not saying it is overall better just did that one task better) Bottom line I’m happy and I’m super excited to see nano banana pro update and also the video models!!!
YouTube
Gemini 4 Argon:十个实用演示
Gemini 4 Argon: 10 incredible AI demos (1M token output in action)
Google Cloud Tech
官方演示涵盖代码迁移、交互式 3D 项目和多模态成果。
查看原文Gemini 4 Argon:从概念图到 3D 大教堂
Gemini 4 Argon and Google Antigravity: From 2D concept to 3D cathedral
Google for Developers and Google Antigravity
智能体结合 Gemini 4 Argon、Antigravity 和 Blender,根据二维参考图重建大教堂。
查看原文Gemini 4 Argon:遗留代码迁移至 Rust
Gemini 4 Argon: Modernizing 40 year code
Google Antigravity and Google for Developers
官方演示使用 Antigravity 和 Gemini 4 Argon,将数十年前的科学计算代码迁移至 Rust。
查看原文