Gemini 4 Argon: our next era of frontier intelligence
- ID: 9e57e69a
- 原文链接: https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon
- 作者: {'platform': 'google', 'author': 'Koray Kavukcuoglu (SVP, Google DeepMind)', 'original_date': '2026-09-30'}
- 日期: 2026-09-30
- 分类: agents
- 来源类型: article
- 标签: gemini, frontier-model, coding-agents, cybersecurity, long-horizon, hn
- 质量评分: 5/5
- 抓取时间: 2026-10-01T04:22:00+00:00
中文导读
Google 发布新一代旗舰模型 Gemini 4 Argon：输出上限从 64K 提到行业领先的 1M token，面向长时程复杂工作流（真实软件工程法律/金融知识工作网络防御）定价 $2/$10 每百万 token，缓存输入享 95% 折扣；DeepSWE v1.1 77.9%CWE-bench v1 68%（并列第一）Zapier AutomationBench 51.3%内部实例重量级：C/C++ 向 Rust 的大规模迁移（最大 800K+ 行 Fuchsia Zircon 内核）libgav1 的 32K 行 SIMD 代码重写后比 Rust 移植版快 2.7 倍agent 集群从机房遥测中自主释放 300+ TiB 内存先行安全措施包括 CoT/行为监控中止机制内部激活监控Gray Swan IPI 基准领先的注入防御，以及高风险训练/评估前封沟 sandbox；首批通过 Fairwind Program 向受信任的网络防御者开放
为什么值得关注
Gemini 4 Argon：1M token 输出上限800K 行内核 C/C++Rust 迁移与自主漏洞修复，首批向网络防御者开放
Grounded in the fetched announcement: output ceiling raised from 64K to 1M tokens, $2/$10 per million tokens with 95%-off cached input, DeepSWE v1.1 77.9% and CWE-bench v1 68% (tied first), plus real migrations of up to 800K+ line C/C++ kernels to Rust with agent fleets.
关键信息
- 标题：Gemini 4 Argon: our next era of frontier intelligence
- 来源：https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon
- 发布时间：2026-09-30
- 关联标签：gemini, frontier-model, coding-agents, cybersecurity, long-horizon, hn
原文摘录
Gemini 4 Argon: our next era of frontier intelligence
发布时间: 2026-09-30
原文链接: https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon
Gemini 4 Argon: our next era of frontier intelligence
Sep 30, 2026
|
9 min read
Gemini 4 Argon delivers frontier performance in complex workflows across real-world software engineering, enterprise knowledge work like legal and finance, and cybersecurity defense.
- * *
[
Koray Kavukcuoglu
SVP, Google DeepMind and Chief AI Architect, Google
](https://blog.google/authors/koray-kavukcuoglu/)
Share
- * *
<audio src="https://storage.googleapis.com/gweb-uniblog-publish-prod/media/tts_audio_86443_umbriel_2026_09_30_21_47_40.mp3" controls></audio>
Listen to article
10:35 minutes
This content is generated by Google AI. Generative AI is experimental
0:00
10:35
Voice Umbriel Speed 1X
Voice UmbrielGacrux
Speed 0.75X 1X 1.5X 2X
Read AI-generated summary
- Google’s new Gemini 4 Argon model brings advanced reasoning to complex, long-horizon professional tasks.
- The model features an industry-leading 1 million token limit for deep, multi-step problem solving.
- It excels at coding, financial research, legal drafting, and autonomous cybersecurity vulnerability patching.
- Argon is currently rolling out to trusted cyber defenders through the Fairwind Program.
- Google is prioritizing safety and rigorous testing before a wider release to the public.
Summaries were generated by Google AI. Generative AI is experimental.
In this article
- * *
Argon will launch at an introductory price 1 of $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at 95% off input token price.
Changing how we work and build at Google
- Quantum algorithmic optimization: Argon is helping our quantum computing researchers optimize the spacetime resources (qubits × gates) of subroutines that bottleneck important applications. In one example, it beat the published baseline by 40% in a matter of minutes.
- Memory efficiency: A team of Argon agents analyzed fleet-wide profiling telemetry to autonomously identify and apply memory optimizations across Google’s data centers, freeing up over 300 TiB of memory once rolled out, with an estimated 500 TiB to 1 PiB in total savings.
Working harder on your most complex problems
Enabling coding and enterprise workflows across domains
Gemini 4 Argon’s capabilities across coding, reasoning, and multimodality and its ability to sustain long, multi-step tasks enable it to excel across a range of enterprise workflows.
Google engineers have been using Argon for their daily tasks, from everyday debugging to large-scale codebase migrations and algorithm designs. It sets a new state of the art on DeepSWE v1.1 (77.9%), which measures a model’s performance in real-world long-horizon software engineering tasks.
Leading in defensive cybersecurity
On CWE-bench v1, which evaluates the model’s ability to remediate security vulnerabilities, Argon ties for first place with a top score of 68%, building on 3.8 Flash Cyber’s frontier performance on CWE-bench v0.
Gemini 4 Argon demonstrates impressive leaps in vulnerability discovery over 3.8 Flash Cyber. For example:
- On Google’
English Summary
Google announces Gemini 4 Argon, its next frontier model for long-horizon professional workflows: output limit raised from 64K to an industry-leading 1M tokens; $2/$10 per million input/output tokens with 95%-off cached input. Benchmarks include DeepSWE v1.1 77.9%, CWE-bench v1 68% (tied first), Zapier AutomationBench 51.3%. Internal deployments include C/C++-to-Rust migrations up to 800K+ lines (Fuchsia Zircon), a libgav1 32K-line SIMD rewrite running 2.7x faster than the Rust port, and agent fleets freeing 300+ TiB of datacenter memory. Safeguards: CoT/action monitoring with execution stops, internal-activation monitoring, leading Gray Swan indirect-prompt-injection robustness, and sealed sandboxes before high-risk training/eval; rollout starts with trusted cyber defenders via the Fairwind Program.
Obsidian Notes
- 内容由
opencli web read拉取页面正文生成，摘录截断自抓取正文。 - 中文导读与价值判断锚定在条目摘要与抓取正文上。