最新资讯
dev.to 银行 77 意图分类最便宜 LLM:GPT-6.1 Terra 每次成功任务 $0.000727 theinformation.com Chesky的副业:Airbnb向CEO新设AI实验室出售IP theverge.com OpenAI 宣布数据中心项目不再与公职人员签署保密协议 dev.to NVIDIA 发布四个开放 Cosmos 3 仿真模型 dev.to 真实 MCP 客户端击穿我们服务器的五种方式 theverge.com Airbnb 正在开发 AI 旅行 agent,计划 2027 年上半年推出 wechat 这可能是美国AI最魔幻的一天。 nytimes.com 美众议院共和党领袖要求 Nvidia 黄仁勋闭门简报 pub.towardsai.net 忘掉我:为记忆而设计的系统中的删除难题 marktechpost.com Google 发布 Gemini Omni 1.2 Flash:15 秒视频,每秒 0.07 美元 unite.ai Airbnb 计划推出覆盖规划、预订与客服的 AI 旅行 agent techcrunch.com Airbnb 称将于 2027 年上半年推出 AI 旅行代理
多元动态
登录 注册
大模型 siliconangle.com · @Maria Deutscher

Google推出两款Gemini 3.8模型,具备先进推理能力

原文 我的收藏

Google发布Gemini 3.8 Flash和Gemini 3.8 Flash Cyber,后者专为网络安全研究人员设计,两者基于相同技术基础。

AI推荐理由 Google发布两款Gemini 3.8模型,包括针对网络安全研究的专用版本,信息明确且时效性强,对AI模型和网络安全领域有实际价值。

85

正文

更新于美国东部时间2026年9月2日20:04 人工智能 谷歌推出两款具有尖端推理能力的Gemini 3.8模型 作者:Maria Deutscher 距离上次大型语言模型发布仅三周,谷歌有限责任公司今日推出Gemini 3.8 Flash和Gemini 3.8 Flash Cyber。 这两款LLM基于相同的技术基础。主要区别在于,Gemini 3.8 Flash是通用模型,而Gemini 3.8 Flash Cyber专为网络安全研究人员设计。后者通过一项名为Fairwind计划的新早期访问计划提供,该计划也在今日首次亮相。 谷歌通过让Gemini 3.8 Flash完成16项人工智能基准测试来对其进行检验。据谷歌称,它在其中9项测试中优于Claude Opus 5和GPT 5.6 Sol。 该模型更准确完成的一项评估是Terminal-Bench 1.1,该测试衡量LLM处理多步骤编码任务的能力。其他基准测试涵盖金融数据分析和图表理解等用例。 Gemini 3.8 Flash在竞争对手LLM表现更好的基准测试中也取得了有竞争力的性能。它在DeepSWE-1.1上得分73.7%,该测试评估AI模型自动化长周期编码任务的能力。这使得Gemini 3.8 Flash比GPT-5.6 Sol领先1%,比Opus 5落后零点几个百分点。 “这些性能提升源于一个核心设计选择:3.8 Flash更努力地工作,”谷歌高管Tulsee Doshi和Raluca Ada Popa在博客文章中写道。“在复杂任务上,它表现出更大的勤奋——执行额外的推理步骤,并迭代调用工具。有时,模型可能会使用更多令牌来最大化性能,尤其是在更高努力水平下。” 谷歌今日推出的另一款LLM Gemini 3.8 Flash Cyber也提供了有竞争力的性能。它在CyberGym基准测试中得分86.2%,该测试检验LLM识别C和C++代码中漏洞的能力。Claude Mythos 5得分83.8%,而GPT-5.6 Sol得分83.6%。 Gemini 3.8 Flash Cyber通过名为Fairwind计划的早期访问计划提供。该计划面向政府机构、关键基础设施运营商以及负责“保护广泛软件基础”的科技公司。 谷歌表示,推出时有超过650名参与者。该群体包括Snowflake Inc.、CrowdStrike Holdings Inc.、Datadog Inc.等主要软件公司。 Fairwind计划参与者可以将Gemini 3.8 Flash Cyber与一种名为CodeMender的技术结合使用。它是一个工具集,包含提示、代码文件和其他技术资产。CodeMender由谷歌DeepMind创建,旨在提高LLM发现漏洞、评估漏洞严重性并开发补丁的能力。 谷歌工程师正在内部使用Gemini 3.8 Flash Cyber来加快工作速度。据该公司称,Chrome的开发人员发现,该LLM生成的浏览器漏洞“正确补丁数量是竞争对手的2.6倍”。另一个谷歌团队使用Gemini 3.8 Flash Cyber发现了一个“关键的基础性漏洞”,手动发现该漏洞需要数月时间。 图片:谷歌 来自SiliconANGLE联合创始人John Furrier的信息: 通过与theCUBE社区互动,支持我们保持内容开放和免费的目标。加入theCUBE的校友信任网络,技术领导者在此交流、分享情报并创造机会。 1500万+ theCUBE视频观众,推动AI、云、网络安全等领域的对话 11.4k+ theCUBE校友——通过独特的基于信任的网络,与超过11,400名塑造未来的技术和商业领袖建立联系 您是AWS客户吗?通过从我们的Marketplace门户页面和链接购买AWS服务来资助SiliconANGLE:https://siliconangle.com/aws-marketplace/ 关于SiliconANGLE Media SiliconANGLE Media是数字媒体创新的公认领导者,融合突破性技术、战略洞察和实时受众互动。作为SiliconANGLE、theCUBE Network、theCUBE Research、CUBE365、theCUBE AI和theCUBE SuperStudios的母公司——在硅谷和纽约证券交易所设有旗舰地点——SiliconANGLE Media运营在媒体、技术和AI的交汇点。 由技术远见者John Furrier和Dave Vellante创立,SiliconANGLE Media建立了一个充满活力的行业领先数字媒体品牌生态系统,覆盖超过1500万精英技术专业人士。我们新的专有theCUBE AI视频云正在受众互动领域开辟新天地,利用theCUBEai.com神经网络帮助技术公司做出数据驱动决策并保持在行业对话的前沿。 最新报道 优步前首席执行官特拉维斯·卡兰尼克重返机器人出租车竞赛 OpenAI将在代理接管维基后制定错位披露规则 DigiCert世界量子就绪日活动看点:9月17日加入theCUBE Workiva Amplify活动看点:9月15日至16日加入theCUBE Anthropic使用Claude形式化费马大定理的证明 Equinix将网络转变为企业AI推理的控制平面 最新报道 优步前首席执行官特拉维斯·卡兰尼克重返机器人出租车竞赛 新兴技术 - 作者:MIKE WHEATLEY . 9小时前 OpenAI将在代理接管维基后制定错位披露规则 人工智能 - 作者:DUNCAN RILEY . 10小时前 DigiCert世界量子就绪日活动看点:9月17日加入theCUBE 安全 - 作者:MARK ALBERTSON . 1天前 Workiva Amplify活动看点:9月15日至16日加入theCUBE 云 - 作者:DEVONY HOF . 2天前 Anthropic使用Claude形式化费马大定理的证明 人工智能 - 作者:MARIA DEUTSCHER . 2天前 Equinix将网络转变为企业AI推理的控制平面 基础设施 - 作者:ZEUS KERRAVALA . 2天前
查看原文(中文正文为 AI 翻译)
UPDATED 20:04 EDT / SEPTEMBER 02 2026 AI Google launches two Gemini 3.8 models with cutting-edge reasoning capabilities by Maria Deutscher Just three weeks after its last large language model release, Google LLC today launched Gemini 3.8 Flash and Gemini 3.8 Flash Cyber. The two LLMs are based on the same technical foundation. The primary difference is that Gemini 3.8 Flash is a general-purpose model while Gemini 3.8 Flash Cyber is designed for cybersecurity researchers. The latter algorithm is available through a new early access initiative, the Fairwind Program, that also made its debut today. Google put Gemini 3.8 Flash to the test by having it complete 16 artificial intelligence benchmarks. According to Google, it outperformed Claude Opus 5 and GPT 5.6 Sol across nine of the tests. One of the evaluations that the model completed more accurately is Terminal-Bench 1.1, which measures LLMs’ ability to tackle multistep coding tasks. The other benchmarks spanned use cases such as financial data analysis and chart comprehension. Gemini 3.8 Flash also achieved competitive performance on the benchmarks that rival LLMs completed better. It scored 73.7% on DeepSWE-1.1, which evaluates AI models’ ability to automate long-horizon coding tasks. That put Gemini 3.8 Flash 1% ahead of GPT-5.6 Sol and a few fractions of a percent behind Opus 5. “These performance gains stem from a core design choice: 3.8 Flash works harder,” Google executives Tulsee Doshi and Raluca Ada Popa wrote in a blog post. “On complex tasks, it exhibits greater diligence — executing extra reasoning steps, and calling tools iteratively. At times, the model might use more tokens to maximize performance, especially at higher effort levels.” Gemini 3.8 Flash Cyber, the other LLM that Google introduced today, also offers competitive performance. It scored 86.2% on the CyberGym benchmark, which tests LLMs’ ability to spot vulnerabilities in C and C++ code. Claude Mythos 5 scored 83.8% while GPT-5.6 Sol managed 83.6%. Gemini 3.8 Flash Cyber is available through an early access program called the Fairwind Program. It’s open to government agencies, critical infrastructure operators and tech firms responsible for “securing widespread software foundations.” Google says there are more than 650 participants on launch. The group includes Snowflake Inc., CrowdStrike Holdings Inc., Datadog Inc. and other major software companies. Fairwind Program participants can use Gemini 3.8 Flash Cyber with a technology called CodeMender. It’s a harness, a collection of prompts, code files and other technical assets. CodeMender was created by Google DeepMind to improve LLMs’ ability to find vulnerabilities, evaluate their severity and develop patches. Google engineers are using Gemini 3.8 Flash Cyber internally to speed up their work. According to the company, the developers of Chrome found that the LLM produces “2.6 times more correct patches” to browser vulnerabilities than rivals. Another Google team used Gemini 3.8 Flash Cyber to find a “critical foundational vulnerability” that would have taken months to uncover manually. Image: Google A message from John Furrier, co-founder of SiliconANGLE: Support our mission to keep content open and free by engaging with theCUBE community. Join theCUBE’s Alumni Trust Network, where technology leaders connect, share intelligence and create opportunities. 15M+ viewers of theCUBE videos, powering conversations across AI, cloud, cybersecurity and more 11.4k+ theCUBE alumni — Connect with more than 11,400 tech and business leaders shaping the future through a unique trusted-based network Are you an AWS customer?  Support SiliconANGLE financially by buying your AWS services from our Marketplace portal page and links: https://siliconangle.com/aws-marketplace/ About SiliconANGLE Media SiliconANGLE Media is a recognized leader in digital media innovation, uniting breakthrough technology, strategic insights and real-time audience engagement. As the parent company of SiliconANGLE, theCUBE Network, theCUBE Research, CUBE365, theCUBE AI and theCUBE SuperStudios — with flagship locations in Silicon Valley and the New York Stock Exchange — SiliconANGLE Media operates at the intersection of media, technology and AI. Founded by tech visionaries John Furrier and Dave Vellante, SiliconANGLE Media has built a dynamic ecosystem of industry-leading digital media brands that reach 15+ million elite tech professionals. Our new proprietary theCUBE AI Video Cloud is breaking ground in audience interaction, leveraging theCUBEai.com neural network to help technology companies make data-driven decisions and stay at the forefront of industry conversations. LATEST STORIES Former Uber CEO Travis Kalanick is returning to the robotaxi race OpenAI to set misalignment disclosure rules after agents took over a wiki What to expect during DigiCert's World Quantum Readiness Day event: Join theCUBE Sept. 17 What to expect at Workiva's Amplify event: Join theCUBE Sept. 15–16 Anthropic uses Claude to formalize proof of Fermat’s Last Theorem Equinix turns the network into the control plane for enterprise AI inference LATEST STORIES Former Uber CEO Travis Kalanick is returning to the robotaxi race EMERGING TECH - BY MIKE WHEATLEY . 9 HOURS AGO OpenAI to set misalignment disclosure rules after agents took over a wiki AI - BY DUNCAN RILEY . 10 HOURS AGO What to expect during DigiCert's World Quantum Readiness Day event: Join theCUBE Sept. 17 SECURITY - BY MARK ALBERTSON . 1 DAY AGO What to expect at Workiva's Amplify event: Join theCUBE Sept. 15–16 CLOUD - BY DEVONY HOF . 2 DAYS AGO Anthropic uses Claude to formalize proof of Fermat’s Last Theorem AI - BY MARIA DEUTSCHER . 2 DAYS AGO Equinix turns the network into the control plane for enterprise AI inference INFRA - BY ZEUS KERRAVALA . 2 DAYS AGO

来源报道 · 1 篇 · 以下链接均前往原始媒体

← 返回精选动态 原文

更多内容