Google's Gemini 4 Carbon Tops AI Index at 59

Key Takeaways

• <ul><li>Google's Gemini 4 Carbon tops the Artificial Analysis Intelligence Index at 59, beating GPT-6 Astra and Claude Opus 5.5</li><li>Gemini 4 Carbon achieves 0% attack success rate on all 15 jailbreak attempts in Gray Swan benchmark</li><li>Model scores 88.2% on SEC-bench Pro, leading all three cybersecurity benchmarks including CWE-bench and ExploitGym</li><li>Gemini 4 Carbon costs $2/$10 per million tokens, 23% cheaper than Claude Opus 5.5 with launch discounts offered</li><li>Vertiv jumps 15% after Q3 earnings beat, reporting $2.21 per share up 78% year over year</li><li>SK hynix says hyperscaler cloud revenue grew 30-80% YoY with backlogs expanding 2-5x</li><li>Gemini 4 Carbon wins Vending-Bench 2 with $16,484 bank balance after 365 simulated days</li><li>Meta launches Waddle World AI pet app in the Philippines on iOS and Android</li><li>FDA clears NeuroDR AI software for 3D artery imaging from CTA scans</li><li>SK Group's AIGEN program offers 3 million won monthly stipend to 4,000 applicants aged 19-34</li></ul>

Gemini 4 Carbon Achieves Perfect Safety Score on Jailbreak Test

Google DeepMind's Gemini 4 Carbon scored a 0% attack success rate on the Gray Swan Indirect Prompt Injection benchmark across 15 attack attempts. The model blocked all automated jailbreak vectors that tried to hide malicious instructions in external data. Competitor models like GPT-6 Astra and Grok 4.6 showed much higher vulnerability rates. This makes Gemini 4 Carbon the most security-hardened large language model in the industry. The result matters because enterprises use AI agents with access to sensitive data and systems.

Google Unveils Gemini 4 Carbon Frontier AI Model

Google announced Gemini 4 Carbon as its new top-ranked AI model, outperforming rivals like GPT-6 Astra and Claude Opus 5.5 on 15 of 17 benchmarks. Carbon scored 59 on the Artificial Analysis Intelligence Index, beating Argon's score of 50. The model costs $2 per million input tokens and $10 per million output tokens. Google says Carbon can work on difficult engineering problems for days at a time. It is currently only available to approved cyber defenders in a limited release.

Gemini 4 Carbon Focuses on Cyber Defense Capabilities

Google DeepMind released Gemini 4 Carbon on October 28, just two weeks after Gemini 4 Argon. Carbon ranks at 59 on the Artificial Analysis Intelligence Index and scored 88.2% on SEC-bench Pro for finding real-world vulnerabilities. The model is being rolled out first to a subset of cyber defenders in the U.S. government's voluntary testing process. DeepMind reports Carbon reduced successful cheating on Terminal-Bench 4.0 to 0.0% and cut high-severity misalignment alerts by almost 80%. The rollout will expand gradually to additional trusted partners.

Gemini 4 Carbon Becomes Top AI Model on Intelligence Index

Google's Gemini 4 Carbon surged to the number one spot on the Artificial Analysis Intelligence Index v5.0 with 59 points, beating Claude Opus 5.5 and GPT-6 Astra. The model scored 70.4% on the Vals Index v2.0 and 88.2% on SEC-bench Pro for cybersecurity defense. Carbon also achieved 77.4% on Agents' Last Exam v1.1 and 81.5% on the OSWorld offline subset. Google designed it as a natively multimodal model with long-context reasoning from 256k to 1M tokens. This ranking solidifies Google's leadership in frontier AI development.

Gemini 4 Carbon Wins Vending-Bench After Year-Long Google Drought

Gemini 4 Carbon topped Vending-Bench 2, developed by Andon Labs, finishing with a leading bank balance of $16,484. The benchmark tasks AI agents with operating a vending machine business across 365 simulated days, managing inventory, suppliers, and pricing. This is the first time in over a year that a Google model reached number one on this benchmark. Carbon demonstrated strong long-horizon reasoning by maintaining context and making compounding financial decisions over months of simulated time. The win signals Google is ready to compete at the frontier of enterprise automation.

Gemini 4 Carbon tops AI intelligence rankings at 23% lower cost

Google's Gemini 4 Carbon scored 59 on the Artificial Analysis Intelligence Index v5, beating Claude Opus 5.5. It costs $7.76 per task compared to $10.04 for Claude Opus 5.5, saving 23%. Google is offering a 50% launch discount and a 95% discount on cached input processing. The model also leads in four of nine Intelligence Index categories and set a new high score on the AA-Omniscience test.

Google's Gemini 4 Carbon detects 88.2% of security flaws

Google introduced Gemini 4 Carbon, an AI model built to find software vulnerabilities and generate security patches. It scored 88.2% on SEC-bench Pro, 71% on CWE-bench, and 54.8% on ExploitGym, ranking first on all three cybersecurity benchmarks. Google said internal teams used the model for days on complex engineering tasks. The model will have introductory pricing of $2 per million input tokens and $10 per million output tokens.

Nvidia partner Vertiv jumps 15% after strong earnings

Vertiv, a data center infrastructure company and Nvidia partner, reported Q3 adjusted earnings of $2.21 per share, up 78% from a year earlier. Its stock rose 15% on Wednesday morning, trading around $240 a share. The earnings beat helped ease concerns about the industry. Vertiv's strong results and guidance drove the positive market reaction.

Vertiv and Amphenol surge while GE Vernova slips on earnings

AI infrastructure stocks reacted differently after earnings from Vertiv, Amphenol, and GE Vernova. All three companies beat Wall Street earnings expectations. Vertiv and Amphenol saw their stocks soar, while GE Vernova slipped. Eric Bleeker of 24/7 Wall St. analyzed the results as a look into the future of the AI boom.

SK hynix says AI investment will continue as cloud sales jump

SK hynix predicted that global big tech companies will keep expanding AI investment. Major hyperscalers saw cloud revenue grow 30% to 80% year over year, driven by AI operations. Cloud backlogs expanded about two to five times from a year earlier. SK hynix noted customers are adjusting the pace of individual projects, but the overall commitment to AI investment remains strong.

FDA clears AI software NeuroDR for better artery imaging

The FDA granted 510(k) clearance to NeuroDR, an AI tool from Deep Reasoning AI. The software removes bone, soft tissue and veins from CTA scans to create 3D images of head and neck arteries. It can show anterior and posterior circulation separately without overlapping anatomy. NeuroDR is the first FDA-cleared product for the company.

Trump plans to ban the term artificial intelligence

President Trump said he wants to ban the term artificial intelligence and replace it with American ingenuity or human ingenuity. He announced this during a speech at a tech conference in Nashville. The White House has not given further details on the plan. Experts are debating whether the ban would slow innovation or improve transparency.

N-able adds AI security tools for MSPs

N-able added AI-assisted vulnerability scanning and patching to its N-central and N-sight platforms. The update automates security for over 900 applications on Windows, iOS and Linux. It aims to reduce exposure time to known threats without adding complexity. This comes about three months after an N-central breach that affected some customers.

Meta launches AI pet app Waddle World

Meta launched Waddle World, an AI pet app, in the Philippines on iOS and Android. The app lets users hatch a virtual pixel pet and interact with it in a 16-bit world. Users can journal with voice or text, play mini-games and receive weekly emails from the pet. Meta says it is not a chatbot, assistant or productivity tool.

AI-found bugs drive corporate security spending

Atomicwork CEO Vijay Rayapati told his team to focus on cybersecurity after serious bugs were found in its code. A machine learning model discovered a critical bug that could have let attackers access company systems. The company is working with a cybersecurity firm to fix the issue and investing in AI-powered security tools. The incident added to a broader trend of companies spending more on AI security.

SK Group launches AI program with 3 million won monthly stipend

SK Group created AIGEN, a two-year AI education program that drew 4,000 people to its first information session. The program is open to adults aged 19 to 34 and does not require coding experience or a specific degree. Participants receive 3 million won per month along with mentoring from Seoul National University's AI Institute. Applications run from the 1st to the 15th of next month, with about 50 people selected by late December to start in March 2027.

AI stocks climb higher after Anthropic IPO filing eases worries

CoreWeave and other AI stocks rose on Tuesday after investors felt more confident about AI spending. Anthropic's IPO filing showed a stronger financial position than many investors had feared. This eased concerns about overspending and high debt levels that had caused tech stocks to fall in recent weeks. CoreWeave's stock climbed 8% following the news.

Sources

• Gemini 4 Carbon Scores A Perfect 0% On Gray Swan IPI Jailbreak Benchmark
• Google announces Gemini 4 Carbon, its new frontier AI model
• Google's Gemini 4 Carbon pushes cyber defense first
• Google’s Gemini 4 Carbon Tops Artificial Analysis Intelligence Index, Becomes Top AI Model In The World
• Gemini 4 Carbon Tops Vending Bench, First Google Model To Top Benchmark In Over A Year
• Gemini 4 Carbon Beats Claude Opus 5.5 On Intelligence While Being 23% Cheaper
• Google's Gemini 4 Carbon AI finds 88.2% of security flaws
• Nvidia Partner Vertiv Soars After Blowout Earnings, Strong Guidance
• 3 AI Stocks Crush Earnings: Vertiv and Amphenol Soar While GE Vernova Slips
• SK hynix forecasts continued AI investment as cloud sales surge up to 80%
• FDA Clears CTA-Based AI Software for Enhanced Visualization of Head and Neck Arteries
• Pres. Trump plans to ban the term 'artificial intelligence.' Here's what he wants to use instead
• N-able arms MSPs with AI endpoint security automation
• Meta has a new AI pet app called Waddle World
• ‘Nobody’s Ready’: AI-Found Bugs Spark a Corporate Security Spending Spree
• SK Draws 4,000 to AI Training Program Paying 3 Million Won a Month
• CoreWeave and other AI stocks rise as Anthropic restores the market’s faith in sustained spending