←新闻详情

Analytics India Magazine更新于 10月30日 14:11

Google 发布 Gemini 4 Carbon,主打网络安全与编码能力

Google Announces Gemini 4 Carbon With Focus on Cybersecurity and Coding

AI 导读

Google 于 10 月 28 日发布 Gemini 4 Carbon,称其为自家最强 AI 模型,重点提升网络安全与编码能力,可识别软件漏洞、评估安全风险并生成修复代码。该模型在 SEC-bench Pro 得分 88.2%、ExploitGym 得分 54.8%、CWE-bench 得分 71.0%,Google 称这些外部安全评测未加护栏。Carbon 建立在两周前发布的 Gemini 4 Argon 之上,初期仅向部分网络安全防御者及 Fairwind Program 可信伙伴开放,尚未公布公开访问日期;介绍期定价为每百万输入 token 2 美元、每百万输出 token 10 美元,之后涨至 4 美元和 20 美元。

推荐理由

精选AI 评分 82

原文列出 Carbon 在网络安全与编码基准上的具体分数和定价,可据此判断其能力定位与成本。

主题

标签

正文

Google announced Gemini 4 Carbon on October 28, describing it as its strongest artificial intelligence model, with improved cybersecurity capabilities and the ability to handle complex, long-running tasks. The model is designed to identify software vulnerabilities, assess security risks and develop code fixes, according to the company.

In a blog post, Koray Kavukcuoglu, Senior Vice President at Google DeepMind and Chief AI Architect at Google, said Carbon builds on Gemini 4 Argon, which launched two weeks earlier. Google said the new model also improves coding, scientific tasks, web development and AI agents that execute multi-step workflows.

Carbon is designed to tackle complex engineering problems that may take days to complete, including identifying the causes of bugs and fixing them across large codebases and low-level systems, the company said.

Google reported that Gemini 4 Carbon scored 88.2% on SEC-bench Pro, a benchmark for identifying vulnerabilities in real-world software. It scored 54.8% on ExploitGym, which evaluates vulnerability exploitation, and 71.0% on CWE-bench, which assesses the ability to fix software weaknesses. Google said these external cybersecurity evaluations were conducted without guardrails.

The company said internal benchmarks showed that Carbon achieved more than twice Argon's success rate in generating secure code. Google’s Chrome team also reported that Carbon performed more than 10 times better than another frontier AI model from a different lab when generating high-quality patches. Google did not identify the competing model or detail the comparison methodology in its announcement.

Artificial Analysis ranked Carbon first on its Intelligence Index version 5.0, with a score of 59. Google also reported an 89-point improvement on WebDev Arena, where Carbon ranked fourth with a score of 1,769.

Google will initially give access to selected cybersecurity defenders, followed by additional trusted partners through its Fairwind Program. The company said it is participating in the US government’s voluntary process for pre-release model testing and will continue refining safeguards before wider availability. It has not announced a specific date for public access.

Gemini 4 Carbon's introductory price is $2 per million input tokens and $10 per million output tokens, matching Argon's introductory rates. Google said the price will rise to $4 per million input tokens and $20 per million output tokens after the introductory period. Cached input tokens receive a 95% discount against the standard input-token price.

Google plans to make Carbon available to developers, businesses and consumers through its products as soon as possible, subject to its phased rollout and continued safety work.

来源

Analytics India Magazine