Google's Gemini 4 Carbon Tops the AI Leaderboard

Google has announced Gemini 4 Carbon, which it calls its strongest model, two weeks after it launched Gemini 4 Argon. In a blog post on 28 October, Koray Kavukcuoglu, chief AI architect at Google, said Carbon has state-of-the-art cybersecurity and "long horizon" capabilities, meaning it can keep working on one task for a long time. Google says internal teams have seen it run for days on hard engineering problems, pushing through dead ends.

How it ranks

On Artificial Analysis's independent leaderboard, Carbon at its high setting scores 59. Claude Opus 5.5 at its max setting scores 58, Opus 5.5 at xhigh and GPT-6 Astra at max both score 57, and GPT-6.1 Sol at max scores 55. The gap at the top is narrow: the first four entries sit within two points.

Artificial Analysis also estimates what each model costs to run its test tasks. It puts Carbon at $7.76 per task, against $10.04 for Opus 5.5 at max, $6.24 for GPT-6 Astra at max and $1.27 for GPT-6.1 Sol at max. On that measure Carbon is not the cheapest strong model, but it is cheaper than the closest scorer.

Security first

Google is leading with cybersecurity. Its own published results put Carbon at 88.2% on SEC-bench Pro, which tests finding vulnerabilities, 54.8% on ExploitGym, which tests exploiting them, and 71.0% on CWE-bench, which tests patching them, according to Collinear's leaderboard. Google says Carbon more than doubles Argon's success rate at writing secure code, and that Chrome's team found it more than 10 times better than "another lab's large frontier model" at producing high-quality patches. Google does not name that lab, and it says these benchmarks were run without guardrails. These are company figures that outside testers have not yet repeated.

Google also reports reliability gains over Argon in its own testing: about half as many unsupported claims, 26% less overreach, 29% less under-service, and a 0.0% attack success rate on Gray Swan's indirect prompt-injection benchmark, against 0.7% for Argon.

More from AI and Tech StakGoogle Pay Arrives in Nigeria With Four BanksGoogle Wallet now takes cards from FirstBank, GTBank, UBA and Zenith. Most supported cards are Mastercard, and phone taps need contactless terminals.

Who can use it, and what it costs

Not the public, yet. Access starts this week with a subset of cyber defenders, then widens to trusted partners through Google's Fairwind Program, and only after that to developers, businesses and consumers "as soon as possible", Google says. It is also taking part in the US government's voluntary pre-release testing of new models.

The introductory price is $2 per million input tokens and $10 per million output tokens, the same as Argon, with cached input 95% cheaper. After the introductory period the price doubles to $4 and $20, according to Google's post and 9to5Google. Constellation Research's Larry Dignan reads the two weeks between Argon and Carbon as a sign that Google is speeding up its release pace.

What is still unknown

• When developers and businesses, including those in Africa, will be able to call Carbon through Google's API or use it in the Gemini app.
• How Carbon performs on security benchmarks run by independent testers rather than Google.
• How partners are chosen for the Fairwind Program, and how long the introductory price lasts.