[{"data":1,"prerenderedAt":14},["ShallowReactive",2],{"artikel-en-berita-google-gemini-4-carbon-keselamatan-siber":3},{"slug":4,"lang":5,"title":6,"description":7,"date":8,"updated":9,"topic":10,"product":9,"cover":11,"readMin":12,"html":13},"berita-google-gemini-4-carbon-keselamatan-siber","en","Google launches Gemini 4 Carbon, its strongest AI model for cybersecurity","Google has announced Gemini 4 Carbon, its strongest model yet with state-of-the-art cybersecurity capability, but rollout is deliberately slow and restricted.","2026-11-02",null,"berita","\u002Fimages\u002Fartikel\u002Fberita-google-gemini-4-carbon-keselamatan-siber.webp",3,"\u003Cp>Google has announced Gemini 4 Carbon, its strongest model yet. It arrives just two weeks after Gemini 4 Argon, a sign of how fast Google is now shipping frontier models. Carbon beats the top models from OpenAI and Anthropic on several cybersecurity benchmarks, but Google is not opening it to everyone just yet.\u003C\u002Fp>\n\u003Ch2>1. What Gemini 4 Carbon is\u003C\u002Fh2>\n\u003Cp>Carbon is Google DeepMind's newest frontier model. According to Google's official announcement on October 28, 2026, Carbon takes the top spot on version 5.0 of the Artificial Analysis Intelligence Index with a score of 59. The model is built for coding, long agentic work, and cybersecurity defense.\u003C\u002Fp>\n\u003Cp>Google says its internal teams have watched Carbon run "for days" on complex multi-step engineering tasks, persisting through dead ends along the way.\u003C\u002Fp>\n\u003Ch2>2. The cybersecurity results\u003C\u002Fh2>\n\u003Cp>This is where Carbon stands out. On cybersecurity benchmarks, Google reports:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>SEC-bench Pro (finding flaws in real software): 88.2%\u003C\u002Fli>\n\u003Cli>ExploitGym (exploitation): 54.8%\u003C\u002Fli>\n\u003Cli>CWE-bench (patching): 71.0%\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>In Google's comparison, Carbon tops GPT-6 Astra from OpenAI as well as Claude Opus 5.5 and Claude Fable 5.1 from Anthropic. SecurityWeek notes these results were obtained with no guardrails in place.\u003C\u002Fp>\n\u003Ch2>3. More honest and more controlled\u003C\u002Fh2>\n\u003Cp>Google also stresses non-technical qualities. In internal testing, Carbon cuts unsupported claims by roughly 2x compared with Argon. It reduced overreach by 26% and under-service by 29%, so it sticks to project scope more reliably.\u003C\u002Fp>\n\u003Cp>On safety, no attack succeeded against Carbon in Gray Swan's indirect prompt injection benchmark, compared with a 0.7% success rate for Argon. Wiz, the Google-owned cloud security firm, says Carbon leads web penetration testing benchmarks.\u003C\u002Fp>\n\u003Ch2>4. A slow rollout, not an open launch\u003C\u002Fh2>\n\u003Cp>Unlike a typical model release, Carbon is not generally available yet. Starting this week, only a small group of cyber defenders gets access. From there, access expands in stages through the Fairwind Program, which connects governments, critical infrastructure operators and major tech platforms with advanced models for cyber defense.\u003C\u002Fp>\n\u003Cp>Google says more than 650 organizations now participate in Fairwind. The company is also part of the US government's voluntary pre-release testing process and is still gathering feedback on guardrails before opening Carbon to developers and consumers.\u003C\u002Fp>\n\u003Ch2>5. Pricing\u003C\u002Fh2>\n\u003Cp>When Carbon reaches developers and businesses, it will launch at the same introductory price as Argon: $2 per million input tokens and $10 per million output tokens. Cached input tokens cost 95% less than the regular input price. After the introductory period ends, the price rises to $4 and $20 per million tokens.\u003C\u002Fp>\n\u003Ch2>What it means for you\u003C\u002Fh2>\n\u003Cp>Most regular users cannot try Carbon today. But the launch hints at where the AI industry is heading: models you can trust with long, critical work, not just models that answer quizzes slightly better.\u003C\u002Fp>\n\u003Col>\n\u003Cli>Expect Carbon to appear in Gemini and the Gemini API in the coming months, at the same introductory price as Argon.\u003C\u002Fli>\n\u003Cli>If you work in cybersecurity or critical infrastructure, consider joining the Fairwind Program for early access.\u003C\u002Fli>\n\u003Cli>These benchmark numbers come from Google itself, so wait for independent testing before making buying decisions.\u003C\u002Fli>\n\u003Cli>For everyday work, Argon and existing models remain enough; Carbon targets heavier tasks.\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Ch2>Sources\u003C\u002Fh2>\n\u003Cul>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Fblog.google\u002Finnovation-and-ai\u002Fmodels-and-research\u002Fgemini-models\u002Fintroducing-gemini-4-carbon\u002F\" target=\"_blank\" rel=\"noopener\">Gemini 4 Carbon: our next frontier model\u003C\u002Fa> (Google Blog)\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Fwww.securityweek.com\u002Fgoogle-says-gemini-4-carbon-leads-in-finding-exploiting-and-fixing-vulnerabilities\u002F\" target=\"_blank\" rel=\"noopener\">Google Touts Powerful Cybersecurity Capabilities in Gemini 4 Carbon Model\u003C\u002Fa> (SecurityWeek)\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Fwww.theverge.com\u002Ftech\u002F1021339\u002Fgoogle-gemini-4-carbon-cybersecurity-frontier\" target=\"_blank\" rel=\"noopener\">Google launches Gemini 4 Carbon with a 'significant leap' in cybersecurity\u003C\u002Fa> (The Verge)\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"https:\u002F\u002F9to5google.com\u002F2026\u002F10\u002F28\u002Fgemini-4-carbon-announcement\u002F\" target=\"_blank\" rel=\"noopener\">Google announces Gemini 4 Carbon with 'significant' coding gains\u003C\u002Fa> (9to5Google)\u003C\u002Fli>\n\u003C\u002Ful>\n",1793599408339]