Choose Rich Live
Tech

Google unveils Gemini 4 Carbon with top benchmark scores and cybersecurity focus

Google has announced Gemini 4 Carbon, framing it as its strongest frontier AI model to date with leading performances across cybersecurity benchmarks and long-horizon coding tasks.

Google benchmark table comparing Gemini 4 Carbon scores against rival AI models across cybersecurity and coding metrics.
Image: @BullTheoryio

Google DeepMind announced the launch of Gemini 4 Carbon on October 28, 2026, positioning the system as its most capable model yet. DeepMind head Demis Hassabis and Google executive Koray Kavukcuoglu described Carbon as state of the art across multiple performance metrics, combining frontier capabilities with deep safety and security research. The release comes two weeks after Google introduced Gemini 4 Argon.[2][4][6][7][10]

According to Google benchmarks and CEO Sundar Pichai, Gemini 4 Carbon ranks first on version 5.0 of the Artificial Analysis Intelligence Index with a score of 59, edging out Anthropic's Claude Opus 5.5 at 58 and OpenAI's GPT-6 Astra at 57. A 17-benchmark evaluation shared by Google showed Carbon achieving top scores in 15 categories, including GraphWalks at 96.2%, RiemannBench at 80.0%, and Agents' Last Exam v1.1 at 77.4%. Rival models retained leads on two tests: Claude Opus 5.5 scored 87.2% on the OSWorld-2.1 offline subset compared to Carbon's 81.5%, while GPT-6 Astra led FrontierSWE 2 at 65.5% over Carbon's 61.2%.[1][3][5][8][9][11]

Google highlighted cybersecurity as Carbon's primary strength, reporting that the model scored 88.2% on SEC-bench Pro for identifying vulnerabilities, 54.8% on ExploitGym for validating exploits, and 71.0% on CWE-bench for patching flaws. Google engineers reported observing Carbon working through complex coding problems over several days, and the Chrome team noted patches generated by Carbon performed more than ten times better than those from another lab's leading model. Early access begins this week, with Google initially limiting availability to select cyber defense teams before initiating a wider rollout.[5][8][11]

Key facts

  • Google unveiled Gemini 4 Carbon on October 28, 2026, two weeks after releasing Gemini 4 Argon.
  • Gemini 4 Carbon scored 59 on the Artificial Analysis Intelligence Index v5.0, ahead of Claude Opus 5.5 at 58 and GPT-6 Astra at 57.
  • According to Google's comparative methodology chart, Gemini 4 Carbon scored highest on 15 of 17 benchmarks, trailing GPT-6 Astra on FrontierSWE 2 and Claude Opus 5.5 on the OSWorld-2.1 offline subset.
  • In cybersecurity evaluations, Gemini 4 Carbon recorded 88.2% on SEC-bench Pro, 54.8% on ExploitGym, and 71.0% on CWE-bench.
  • Google stated that Carbon achieves more than double the rate of Gemini 4 Argon at generating secure code.
  • Early access to Gemini 4 Carbon starts the week of its announcement, initially restricted to a limited group of cyber defenders.

Sources · 11 sources

  1. PO

    Polymarket@PolymarketPost on X ·

    BREAKING: Google unveils Gemini 4 Carbon, its most powerful AI model yet, surpassing Claude Opus 5.5 & GPT-6 Astra on the Artificial Analysis Intelligence Index.

    Open source
  2. TE

    TechmemeArticle ·

    Google unveils Gemini 4 Carbon, its "strongest model with state-of-the-art cybersecurity" capabilities, ranked #1 on Artificial Analysis' new Intelligence Index (Koray Kavukcuoglu/Google) Koray Kavukcuoglu / Google : Google unveils Gemini 4 Carbon, its “strongest model with state-of-the-art cybersecurity” capabilities, ranked #1 on Artificial Analysis' new Intelligence Index — Gemini 4 Carbon is our strongest model with state-of-the-art cybersecurity and long horizon capabilities.

    Open source
  3. AC

    Andrew Curran@AndrewCurran_Post on X ·

    Gemini 4 Carbon benchmarks; https://t.co/vZGjpmyfh6

    Open source
  4. DH

    Demis Hassabis@demishassabisPost on X ·

    ⚡Iterating at lightning speed! Gemini 4 Carbon is SOTA across key benchmarks and combines frontier capabilities with deep safety & security research. https://t.co/TC8Y8ziPOI

    Open source
  5. WS

    Wall St Engine@wallstenginePost on X ·

    Google just unveiled Gemini 4 Carbon. Its most advanced AI model yet, with a major focus on cybersecurity and long-running engineering tasks. Google says its engineers have seen Carbon work through complex coding problems for days, pushing past dead ends without getting stuck. According to Google's benchmarks, Carbon scores 59 on the Artificial Analysis Intelligence Index, ahead of Claude Opus 5.5 and GPT-6 Astra. Its cybersecurity results are particularly notable: • 88.2% on SEC-bench Pro for finding vulnerabilities • 54.8% on ExploitGym for validating exploits • 71.0% on CWE-bench for patching security flaws Google also reports more than 2x Argon's success rate at generating secure code. Carbon will initially be available to a limited group of cyber defenders before a broader rollout.

    Open source
  6. ㆅ

    ㆅ@howfxrPost on X ·

    Google announces Gemini 4 Carbon, its latest frontier AI model, with a focus on advanced coding, long-running agentic tasks and cybersecurity. Google says Carbon is its strongest model to date, building on Gemini 4 Argon, launched two weeks ago. 🧵 https://t.co/UXjsmTs47Z

    Open source
  7. CB

    Crypto BriefingArticle ·

    Google DeepMind announces Gemini 4 Carbon with cybersecurity focus Gemini 4 Carbon's focus on cybersecurity and long-horizon tasks could redefine industry standards, impacting future AI deployment strategies. The post Google DeepMind announces Gemini 4 Carbon with cybersecurity focus appeared first on Crypto Briefing .

    Open source
  8. BT

    Bull Theory@BullTheoryioPost on X ·

    BREAKING: Google $GOOGL has launched its smartest AI model yet, Gemini 4 Carbon. On the Artificial Analysis Intelligence Index, which ranks the top AI models, it scored 59. That puts it at #1, ahead of Claude Opus 5.5 at 58 and OpenAI's GPT-6 Astra at 57. Google says it can keep working on a complex coding problem for days. But its biggest strength is cybersecurity. It was tested on the three main steps of protecting software: 1. Finding security flaws: it found them in real software 88.2% of the time 2. Checking if those flaws can be used to attack a system: it scored 54.8%, against 42.4% for GPT-6 Astra and 34.5% for Claude Opus 5.5 3. Fixing them: it wrote working fixes 71.0% of the time Google's Chrome team says it made patches more than 10 times better than another lab's top model. Google says it is so good at finding flaws and attacking them that only few cybersecurity groups are allowed to use it for now.

    Open source
  9. SI

    Solid Intel 📡@solidintel_xPost on X ·

    INTEL: Google unveils Gemini 4 Carbon, an AI model Pichai says ranks #1 on the Artificial Analysis Intelligence Index

    Open source
  10. NE

    NeowinFeed@NeowinFeedPost on X ·

    Google has unveiled Gemini 4 Carbon just two weeks after releasing Argon, escalating the frontier artificial intelligence race against Anthropic and OpenAI. #Google #Gemini4Carbon #Gemini https://t.co/86x3kF88Vx

    Open source
  11. CB

    Crypto Briefing@Crypto_BriefingPost on X ·

    🚀 NEW: Google introduces Gemini 4 Carbon, its latest frontier AI model, ranking #1 on Artificial Analysis' Intelligence Index v5.0 with a score of 59. The model also leads cybersecurity benchmarks, with early access starting this week. https://t.co/4PqigK2AGv

    Open source