Gemini 4 Carbon: Release Date, Price and Who Can Use It

Disclosure: We may earn a commission when you buy through links on this page, at no extra cost to you. How we make money.

Is Gemini 4 Carbon available?

Not to most people. Google’s announcement says Carbon goes first to “a subset of cyber defenders,” with access phasing out to more trusted partners through its Fairwind Program. Developers, businesses and consumers come after that, “as soon as possible.”

We checked the places where a public launch would show up:

• Gemini API: Carbon isn’t on the models page, the pricing page or the changelog.
• Gemini app plans: Google’s subscriptions page lists Argon and doesn’t mention Carbon.

One source muddies this. Artificial Analysis’s Carbon page says “No publicly available API” in its charts, while its FAQ says Carbon is available through one provider. Its numbers are labelled as based on Google’s API, but the page doesn’t explain how it got access. Going by Google’s own pages, you can’t sign up and call Carbon today.

Gemini 4 Carbon release date: what can we expect?

Google hasn’t given one, so any exact date you see is a guess. The one useful comparison is Carbon’s older sibling:

• Argon was announced on September 30 for “trusted cyber defenders.”
• It became generally available in the API on October 13, 13 days later.

The same gap from October 28 would land around November 10. Treat that as arithmetic, not a forecast. One earlier launch can’t establish a pattern. Watch the Gemini API changelog, since that’s where Argon’s release appeared.

Can your company get early access?

Only if it fits a narrow profile. The Fairwind Program page prioritizes governments and national cyber authorities, critical infrastructure operators, and core technology platforms. Academic labs doing defensive benchmarking can also apply.

You request access through your Google Cloud account team, and Google runs background checks on applicants. The page describes these terms for Argon and doesn’t mention Carbon, so ask your account team whether joining would include it. For a typical small business or solo developer, this isn’t a realistic route.

Gemini 4 Carbon price, and what it could really cost

Google says Carbon will launch at Argon’s introductory price, with cached input tokens 95% cheaper than regular input. A later price applies after the introductory period.

Here’s what that means for a sample monthly workload of 10 million input tokens and 2 million output tokens (our example, not a typical bill):

• Introductory price: $20 + $20 = $40
• Later price: $40 + $40 = $80
• Introductory, with 8 million of the input cached: $0.80 + $4 + $20 = $24.80

Two details are easy to miss. Argon’s introductory price ends on November 30 and doubles on December 1. Carbon’s announcement gives no end date, so don’t build a budget on $2 and $10 lasting.

The other is token use. Artificial Analysis calls Carbon “very verbose,” at about 200 million output tokens on its test set against a 70 million median. A low price per token doesn’t help much if the model writes far more of them, so test your own workload once you have access.

Gemini 4 Carbon vs Argon, Opus 5.5 and GPT-6

Google says Carbon ranks first on version 5.0 of the Artificial Analysis Intelligence Index, with a score of 59. The leaderboard confirms that, and it’s the most useful number available because Artificial Analysis says it measures models independently.

Source: Artificial Analysis, read October 31, 2026. The Carbon page lists $4 and $20 as its price and doesn’t say which rate the cost figure uses, so treat these costs as rough.

Our reading of this table:

• Carbon vs Opus 5.5 is close. One point separates the top settings, which is no reason to switch a working setup by itself. The Opus rows also carry a “with fallback” label that Carbon’s doesn’t, so they aren’t measured under identical conditions.
• Carbon vs Argon is the bigger story. It scores 9 points higher at nearly the same listed cost per task. If that holds at launch, Argon users on API work have a clear reason to test it.
• “#1” doesn’t mean best value. GPT-6.1 Sol at max scores 55 for $1.27 a task. If your work is high-volume, that gap in cost could matter more than 4 index points.

Compare scores only within one index version. Older articles quote different figures for the same models, including a higher score for Argon, because the index has changed.

Google’s own claims

Google’s post also reports results we can’t confirm independently:

• Security benchmarks: SEC-bench Pro 88.2% for finding vulnerabilities, ExploitGym 54.8% for exploitation, CWE-bench 71.0% for patching. Google says these were “evaluated without guardrails,” so public users may see different behavior.
• Secure code: “more than 2x uplift” over Argon in success rate, on an internal test.
• Web development: up 89 points to #4 on WebDev Arena, at 1768. So it isn’t first everywhere.
• Prompt injection: 0.0% on Gray Swan’s test, against 0.7% for Argon.

Why does Carbon go to cyber defenders first?

Google’s scores show a model that can find and exploit software flaws, not just patch them. Its post says the external security benchmarks ran without guardrails, and that it’s taking part in the US government’s voluntary pre-release testing. Our reading: letting defenders in first gives them a head start before a wider release. Google followed the same order with Argon.

What should you use until Carbon opens up?

• Gemini app users: Argon comes with Google AI Pro and Ultra, and the free plan doesn’t list it. Don’t upgrade expecting Carbon, since Google hasn’t said which plans will get it. Our Google AI Pro vs ChatGPT Plus comparison can help if you’re choosing a plan.
• API developers: use gemini-4-argon, and budget for its price doubling on December 1. Keep the model name in one config setting so switching is easy later.
• High-volume or cost-sensitive work: compare GPT-6.1 Sol first, going by the cost-per-task figures above.
• Security teams at eligible organizations: ask your Google Cloud account team about Fairwind.
• Everyone else: there’s nothing to do yet. See our GPT-6 vs Gemini 4 comparison and ChatGPT vs Claude guide for tools you can use now.

Sources

• Google: Gemini 4 Carbon announcement (October 28, 2026)
• Google: Gemini 4 Argon announcement (September 30, 2026)
• Gemini API models, pricing and changelog
• Google DeepMind: Fairwind Program
• Google: Gemini subscriptions
• Artificial Analysis: Gemini 4 Carbon and model leaderboard

Pages checked October 31, 2026.

Frequently Asked Questions