Gemini 4 Carbon: what's new and when you can use it

Short answer: Gemini 4 Carbon is Google's new frontier AI model, announced on October 28, 2026, and it currently tops the Artificial Analysis Intelligence Index. It is built for long-running coding and cybersecurity work, but you can't use it yet unless you're one of the cyber defenders Google is letting in first. Everyone else, including Gemini app subscribers and developers, keeps using Gemini 4 Argon until Google widens access, and it hasn't given a date.

If you use Gemini, build on Google's models, or are weighing Gemini against Claude and ChatGPT, here's what Carbon is, what Google says it does better, who can get it and what to do in the meantime. We've used Google's announcement, its model page and its published evaluation notes, and we flag where figures are Google's own.

What Google announced

Google announced Gemini 4 Carbon on October 28, 2026, calling it "our strongest model with state-of-the-art cybersecurity and long horizon capabilities". It arrives two weeks after Gemini 4 Argon, which Google made available to developers, businesses and Gemini subscribers on October 13, 2026.

The headline claims are:

• Top of the leaderboard. Carbon ranks first on version 5.0 of the Artificial Analysis Intelligence Index with a score of 59, one point ahead of Anthropic's Claude Opus 5.5.
• Long-horizon work. Google says its own teams have seen Carbon run for days on complex, multi-step engineering tasks, such as tracking down and patching subtle bugs across large codebases.
• Cybersecurity. Google reports leading scores on benchmarks for finding vulnerabilities, checking whether they can be exploited and patching them.
• Better behaviour. Google says Carbon makes roughly half as many unsupported claims as Argon, is less likely to cut corners on tests, and resisted every attack in a third-party prompt injection benchmark.

Who can use Gemini 4 Carbon right now

Almost nobody, for now. Google says it is taking "a deliberate approach to rollout" and is part of the U.S. government's voluntary pre-release testing process. Starting the week of the announcement, Carbon goes to "a subset of cyber defenders", with access then phased to more trusted partners through Google DeepMind's Fairwind Program.

Fairwind is an application-only programme for organisations such as governments, national cyber authorities, critical infrastructure operators (healthcare, telecoms, energy, finance) and core technology platforms. Google vets applicants, limits use to defensive and research tasks, and requires controls such as phishing-resistant multi-factor authentication. Individuals and most companies won't qualify.

Google says it will bring Carbon to "developers, businesses, and consumers via our product surfaces as soon as possible", but it hasn't said when, or which Gemini plans will get it. Until then:

• In the Gemini app, Gemini 4 Argon is the top model, offered on the Google AI Pro and AI Ultra plans (see Google's Gemini subscriptions page for what each plan includes in your country).
• For developers, Argon is available on the paid tier of the Gemini API in Google AI Studio, and Google AI Ultra subscribers can use it in Google Antigravity, Google's agent-first development platform.
• In third-party coding tools, GitHub Copilot lists Gemini 4 Argon among its supported models and Cursor offers Argon too. Neither listed Carbon when we checked on October 29, 2026.

What's new in Carbon

Carbon isn't a new product you open; it's the model that will eventually sit behind Gemini and Google's developer tools. Google highlights three areas.

Longer, harder coding tasks

Google says Carbon is "built for sustained agentic workflows", meaning it can keep working through a long job, recover from dead ends and finish multi-step plans. It also reports better output for full-stack web development, with Carbon climbing to fourth place on the WebDev Arena leaderboard. Google says Carbon cut both "overreach" (doing more than asked) and "under-service" (doing less than asked) compared with Argon, which matters if you hand an agent a task and walk away.

Cybersecurity defence

This is the main reason for the restricted launch. Google reports that Carbon scored 88.2% on SEC-bench Pro (finding real vulnerabilities), 54.8% on ExploitGym (checking whether a flaw can be exploited) and 71.0% on CWE-bench (writing working fixes). Google also says its Chrome team found Carbon far better than another lab's large frontier model at producing high-quality patches. The same skills that help defenders could help attackers, which is why Google is starting with vetted security teams.

Safety and honesty

Google says Carbon reduced successful cheating on the Terminal-Bench 4.0 coding benchmark to zero, cut high-severity misalignment alerts in its internal coding environments by almost 80%, and now flags discrepancies to the user before acting rather than guessing. On Gray Swan's indirect prompt injection benchmark, where Argon had a 0.7% attack success rate, Google says Carbon blocked every attempt. These are Google's own evaluations; independent testing will matter once more people can use the model.

How Carbon compares with Claude and GPT

Google's Gemini model page compares Carbon with Argon, OpenAI's GPT-6 Astra and Anthropic's Claude Opus 5.5. A selection of rows:

Two things to keep in mind. First, Carbon doesn't win every row: on Google's own chart, GPT-6 Astra leads FrontierSWE 2 and Claude Opus 5.5 leads OSWorld-2.1 for computer use. Second, Google's evaluation notes say scores for non-Gemini models mostly come from the providers' self-reported numbers, and some Gemini scores were run by Google itself. Benchmarks are a guide, not a verdict on which model will suit your work.

The bigger practical difference today is availability. Claude Opus 5.5 and OpenAI's latest models are available to paying users now, while Carbon is limited to vetted security partners. If you need a top model this week, Carbon isn't an option yet.

What it means if you're choosing AI tools

• Gemini app users: nothing changes today. On Google AI Pro or Ultra, Argon is the best model you can use. Wait for Google's announcement of wider access before paying more on the assumption that Carbon is included.
• Developers on the Gemini API: build and test on Argon now. Google says Carbon will launch at the same introductory API price as Argon, which suggests switching should be straightforward when it opens up. Check Google's live pricing before you budget, because Google says the introductory price won't last.
• Teams using coding agents: if you use GitHub Copilot or Cursor, you can already pick Gemini 4 Argon alongside Claude and GPT models, which makes it easy to test Google's model without changing tools. See Cursor vs GitHub Copilot or Claude Code vs Cursor if you're choosing a coding tool.
• Security teams: if your organisation runs critical services, the Fairwind Program is the way in. If you're not eligible, Google points to its CodeMender code security agent running on publicly available models.

The release also shows how fast the frontier is moving. Google shipped Argon and Carbon two weeks apart, and the top of the Artificial Analysis leaderboard is now separated by a point or two. For most people, that's an argument for tools that let you switch models, such as GitHub Copilot, Cursor or Claude Code alongside a Gemini subscription, rather than locking into one provider. Our Claude vs Gemini and ChatGPT vs Gemini comparisons cover the everyday differences between the apps.

FAQ

What is Gemini 4 Carbon?

Gemini 4 Carbon is Google's newest frontier AI model, announced on October 28, 2026. Google describes it as its strongest model, with state-of-the-art cybersecurity and long-horizon capabilities, and says it ranks first on the Artificial Analysis Intelligence Index v5.0 with a score of 59.

Can I use Gemini 4 Carbon in the Gemini app?

Not yet. Google is giving Carbon first to a group of cyber defenders and then to more trusted partners through its Fairwind Program. It says Carbon will come to developers, businesses and consumers as soon as possible but hasn't given a date. Gemini 4 Argon remains the top model in the Gemini app for now.

What's the difference between Gemini 4 Carbon and Gemini 4 Argon?

Carbon is the newer and, on Google's benchmarks, stronger model, with the biggest gains in long-running coding tasks and cybersecurity. Argon launched on October 13, 2026 and is widely available today in the Gemini app on AI Pro and Ultra, the Gemini API and tools such as GitHub Copilot and Cursor.

Is Gemini 4 Carbon better than Claude and ChatGPT?

On some benchmarks. Google's chart shows Carbon ahead of Claude Opus 5.5 and GPT-6 Astra on the Artificial Analysis index and several coding and security tests, but behind GPT-6 Astra on FrontierSWE 2 and behind Claude Opus 5.5 on OSWorld-2.1. Competitor figures are mostly self-reported by the other labs, and Carbon isn't publicly available yet, so it can't be tested independently by most users.

How much will Gemini 4 Carbon cost?

Google says Carbon will launch on the Gemini API at the same introductory price as Argon, $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 after the introductory period. Consumer plan details haven't been announced, so check Google's live pricing pages when it becomes available.