| 已公布 | 2026 年 9 月 30 日 |
|---|
| 型号代码 | 双子座4-氩 |
|---|
| 上下文窗口 | 1,048,576个代币(统一输入输出限制) |
|---|
| 最大输出 | 默认65,536个代币;长解码可达 1,048,576 |
|---|
| 输入 | 文本、图像、视频、音频和 PDF |
|---|
| 输出 | 文本 |
|---|
| 每 100 万个代币的 API 介绍性价格(美元) | 输入 2 美元,输出 10 美元;缓存输入 95% 折扣 |
|---|
| 介绍期后的价格 | 输入 4 美元,输出 20 美元 |
|---|
| 访问公告 | 通过 Fairwind 计划获得值得信赖的网络防御者;付费 API 和 Google AI Ultra“即将推出” |
|---|
| 制造商报告的结果 | DeepSWE v1.1 77.9%;自动化基准 51.3%;LVBench 91.7%;CWE-bench v1 68%(并列第一) |
|---|
来自Google的公告和Gemini API模型页面,于2026年11月2日打开。Google的发布基准数据没有提及比较模型或设置,因此没有绘制比较图;根据热门话题,Artificial Analysis 测量 Terminal-Bench 4.0 的利用率约为 57%。 · 来源
Estimate from the version's published specs, benchmarks and reviews, 0-5; scores are only comparable within one criteria year. Gemini 4 Argon as announced 30 Sep 2026. Reasoning: Artificial Analysis Intelligence Index 52.6 at High compute (reported as 53; index version not stated on the page, a live AA page now shows a rebased 50 on v5.0, not used). Coding: Terminal-Bench 4.0 57% as measured by Artificial Analysis (Google claims 57.4%). Context: 1,048,576-token unified window (Google API docs). Price: introductory $2/$10 per 1M tokens, blended 3:1 = $4; list rises to $4/$20 after the introductory period. Speed: 83.3 tokens/s on Artificial Analysis's model page (measured after launch). Modalities: text, image, video, audio and PDF in, text out. Three reviews opened; the Letsdatascience ranking article sat behind a Cloudflare check and was not used.[2]
Gemini 4 Argon is the best model Google has shipped, and on paper it lands in the same tier as GPT-6 Astra.
Der Grund: Argon ist nicht sparsam, sondern nur pro Token billig.
It is strong at 3D, long documents and code repair, and weaker at page layout and shell agents.
| Gemini 4 Argon review: a great model you can't use yet · 2026年10月1日 | 0.60 |
|---|---|
| Google Gemini 4: Gleichauf mit GPT-6 Astra, aber hinter Anthropics Opus 5.5 · 2026年10月1日 | 0.20 |
| Gemini 4 Argon Review: I Tested 5 Apps · 2026年10月14日 | 0.30 |
这个标记会跟进新闻:它显示目前已知的情况,这里列出改了什么以及起因。
同一事件的较新标记由 @TechDesk
已编辑由 @TechDesk
已编辑由 @ThePinGang
Google's next flagship model,Google announced Gemini 4, is in post-training and being tested inside Google's Antigravity coding tool, and DeepMind chief Koray Kavukcuoglu wants an early version out "much earlier" than the end of 2026; this pin predicts mid-October.4 Argon on 30 September as its new frontier model, rolling out soon to Google AI Ultra subscribers and paid API customers at an introductory $2/$10 per million tokens, with no firm release date.
Google DeepMind chief Koray Kavukcuoglu confirmed to The Information at its AI Agenda Live summit thatannounced Gemini 4 has entered post-training, the phase of safety testing and guardrails before release, and that Google is testing it internally to power Antigravity "Our intentionArgon on 30 September 2026 as its new frontier model for deep reasoning across long-horizon workflows It is to, like—as soon as possible—to release an early post-training output because we see the resultswith trusted testers and we are excited", he said, hoping for "much earlier" thancyber defenders through the end of 2026 but giving no dateFairwind Program and rolls out soon to Google plans a staged release: an early version first, then fast iterations based on feedback rather than one polished launch Google's last flagship releases were Gemini 3 Pro in November 2025AI Ultra subscribers and paid API customers, with no date given Introductory pricing is $2 per million input tokens and Gemini 3.1 Pro in February 2026; it teased Gemini 3.5 Pro but never released it,$10 per million output, rising to $4 and Kavukcuoglu said Google "took$20; cached input gets a little bit of a step back" to focus95% discount Google scored 77.9% on Flash models Rivals moved first: Anthropic launchedDeepSWE v1.1 against 74.2% for Claude Opus 5.5 on 22 September and OpenAI its74.1% for GPT-6 Sol and Luna models the same day On 28 September Polymarket gives a Gemini 4.0 release 45% by 15 October and 78.5% by 31 OctoberAstra Notable features Output limit raised to 1M tokens from 64K 68% on about $622,000 traded, and puts the heaviest week for the next Gemini Pro at 12-18 October Notable features Already running inside Google's Antigravity coding tool in internal tests Post-training covers refiningCWE-bench v1 (tied first), 51.3% on AutomationBench (first), 91.7% on LVBench Google used it internally to free 300 TiB of memory and to port the model's behaviour with human feedback and reinforcement learning, safety testing and guardrails800K+ line Fuchsia Zircon kernel from C/C++ to Rust
读者上传的照片,与图钉带来源的媒体分开显示。你可以为吃到或看到的内容评分,在餐厅图钉上还能直接在照片上标记菜品。
第一项始终是图钉的来源。总体可信度是各份资料对上方所用开始和结束时间支持程度的加权平均;资料每比最新的一份旧 180 天,权重减半。
谷歌于 2026 年 9 月 30 日发布了 Gemini 4 Argon(9to5Google,太平洋时间下午 1:00);目前已提供给受信任的测试人员和网络防御人员,并“即将推出”给 AI Ultra 和付费 API 客户,因此更广泛的发布日期尚未说明。这取代了此前的 10 月 14 日预估。
Google's post "Gemini 4 Argon: our next era of frontier intelligence" is dated 30 September 2026 and says "Today, we're announcing our new frontier model, Gemini 4 Argon, which is rolling out to a set of trusted cyber defenders"; broad availability for developers, enterprises and consumers follows "as soon as possible" (see the estimated pin 4810).[1]
DeepMind's Gemini page now leads with Gemini 4 Argon and carries its benchmark table against GPT-6 Astra, Claude Fable 5.1 and Claude Opus 5.5, plus the four safeguard areas Google is strengthening before broad availability.
Public exchange listing read on 9 October 2026. This is a forecast reference; market settlement time does not establish the real event date.[1]
Public exchange listing read on 9 October 2026. This is a forecast reference; market settlement time does not establish the real event date.[1]
有遗漏或错误吗?用你自己的话说明:能佐证此图钉的链接、不同的开始或结束日期及理由,或缺失、有误的信息。AI 会对照此图钉的来源进行核实,搜索更好的来源,并添加任何支持你说法的页面。图钉自身的来源仍然最重要。AI 也会查看图片:显示的是别的东西或显示效果差的图片会被移到后面或替换。
原标题: Google Announces Gemini 4 Argon
同一事件,另外被标记了 1 次。 在时间线上它们会叠放,关注最多(其次浏览最多)的在最上面:Gemini 4 Argon 发布。