Gemini 4 Carbon: Google's new top AI model goes to cyber defenders first

What Google announced. Gemini 4 Carbon arrives two weeks after Gemini 4 Argon. Google calls it its strongest model, focused on cybersecurity and tasks that run a long time. The company says its own teams have seen the model work for days on hard multi-step jobs, such as finding and fixing subtle bugs in large codebases.

The numbers. Carbon scores 59 on the Artificial Analysis Intelligence Index v5.0, the top spot. The Verge says Claude Opus 5.5 and GPT-6 Astra trail behind. On security tests Google cites 88.2% on SEC-bench Pro (finding vulnerabilities), 54.8% on ExploitGym (checking whether a flaw can be exploited) and 71.0% on CWE-bench (patching). On WebDev Arena it gained 89 points to reach fourth place. These are tests the company chose and published itself.

Who gets access. Starting this week, a subset of cyber defenders gets Carbon, then trusted partners through the Fairwind Program. Developers, businesses and consumers come later; Google only says "as soon as possible". The company also takes part in the US government's voluntary pre-release model testing.

What it costs. The introductory price matches Argon: $2 per million input tokens and $10 per million output tokens, with cached input 95% cheaper. A token is a small piece of text, roughly part of a word. After the introductory period the price becomes $4 and $20.

Safety. Google says Carbon makes unsupported claims about half as often as Argon and resisted every attack in Gray Swan's prompt injection test.

What we don't know. When Carbon becomes available to everyone and how long the introductory price lasts. Independent checks of these results are still few.

Confirmed facts

• On 28 October 2026 Google announced Gemini 4 Carbon and called it its strongest model.^[1][2]
• Carbon ranks first on the Artificial Analysis Intelligence Index v5.0 with a score of 59.^[1]
• On that index Claude Opus 5.5 and GPT-6 Astra trail Carbon.^[2]
• Security benchmark results: SEC-bench Pro 88.2%, ExploitGym 54.8%, CWE-bench 71.0%.^[1]
• Carbon comes two weeks after Gemini 4 Argon.^[1][2]
• From this week a subset of cyber defenders gets access, then trusted partners via the Fairwind Program; there is no date for developers and consumers.^[1][2]
• Introductory price is $2 per million input and $10 per million output tokens; after the introductory period, $4 and $20.^[1]
• Google takes part in the US government's voluntary process for pre-release model testing.^[1]
• Google says Carbon cuts unsupported claims by about half compared with Argon.^[1][2]
• On WebDev Arena the model gained 89 points to rank fourth (1768).^[1]

Implications

• ConfirmedDevelopers cannot use Carbon yet: access is limited to a small group.
• Our readingThe pace of releases puts pressure on rivals OpenAI and Anthropic and may push AI prices down faster.
• Our readingThe focus on cyber defense suggests such models are becoming tools for security teams.

Still unknown

• When Carbon opens to developers, businesses and consumers.
• How long the introductory price lasts.
• Whether independent tests confirm Google's numbers.

Who it affects

United States, Worldwide. Developers, security teams and people who follow the AI race and Alphabet shares

Sources

1. Google: Gemini 4 Carbon: our next frontier model October 28, 2026 · official
2. The Verge: Google launches Gemini 4 Carbon with a 'significant leap' in cybersecurity October 28, 2026 · media