What Google's Gemini 4 Carbon is

Carbon is Google's latest frontier model, and Pichai kept the reveal casual. "I did say we'd be releasing our models periodically... introducing Gemini 4 Carbon, our next frontier model," he posted, adding that coming so soon after Argon it "shows the pace at which we're pushing the frontier." The main claim is simple. Carbon ranks first on the newly updated v5.0 of Artificial Analysis's index, the closest thing the field has to an overall scoreboard. On that index it scores 59, just ahead of Claude Opus 5.5 at 58, GPT-6 Astra at 57, and its own two-week-old sibling Argon at 50.

Sundar Pichai@sundarpichai
I did say we'd be releasing our models periodically...introducing Gemini 4 Carbon, our next frontier model. Following Argon's release just two weeks ago, Carbon shows the pace at which we’re pushing the frontier, it ranks #1 on the recently updated v5.0 of @ArtificialAnlys Intelligence Index at 59. It also leads across key cybersecurity evaluations scoring 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching. We’re seeing significant improvements in long-horizon tasks even compared to Argon, and we’ve strengthened frontier safeguards. As we did with Argon, we’re going to roll it out gradually + responsibly. Some trusted cyber defenders will get access starting this week. We'll be expanding it to more partners through our Fairwind Program, ahead of wider availability soon, so hang tight!
View on X

Why Gemini 4 Carbon's cybersecurity scores stand out

The gaps get wider on specific tasks. Carbon runs away with long-context work, scoring 96.2% on a test that tracks reasoning across up to a million tokens, and it tops the agent and automation benchmarks too. The headline, though, is security. Carbon sweeps that whole section, leading on CWE-bench, ExploitGym and a SEC-bench Pro score of 88.2%. That edge matters more than a leaderboard place right now. It lands just after the PHP exploits that pulled websites offline worldwide and put AI-driven attacks at the center of the conversation.

Google Gemini 4 Carbon Explained as the New Frontier Model Tops AI Benchmarks on Cybersecurity - Logiciel Solutions Read more: Google Gemini 4 Carbon Explained as the New Frontier Model Tops AI Benchmarks on Cybersecurity

Where rival models still beat Gemini 4 Carbon

Google did not bury the losses, since it published the full comparison. GPT-6 Astra still beats Carbon on one of the tougher software-engineering tests, FrontierSWE 2, scoring 65.5% to Carbon's 61.2%. Claude Opus 5.5 leads clearly on OSWorld, the computer-use benchmark, at 87.2% against Carbon's 81.5%. So the top of the table is less a throne than a scrum, with four or five models separated by a point or two and swapping wins depending on the task.

The release also backs up the pace Pichai pointed to. Carbon arrived two weeks after Argon, with GPT-6 Astra and the Claude 5.5 line all shipping inside a few months, so the number one ranking now changes hands almost as fast as the models land. The cybersecurity result carries its own catch too. A model that is the best at finding and fixing software flaws is, by the same skill, very good at finding them to exploit.