Google ships Gemini 4 Carbon at $2 input, until Nov. 30

The window is the news.

Google released Gemini 4 Carbon to paying developers and enterprises on Tuesday, and for the next three weeks it costs exactly what Gemini 4 Argon does. Then the price doubles.

Carbon is out of the defenders' preview

Google says Carbon is available to all paying developers and enterprises through Google AI Studio and Gemini for business, and in the Gemini API paid tier in AI Studio and Android Studio.

It is also rolling out to Google AI Ultra subscribers in the Gemini app, AI Mode in Search and Google Antigravity.

Two weeks ago, Carbon went only to a set of cybersecurity defenders through Google's Fairwind Program, as TopFive reported then.

The introductory price matches Argon: $2 per million input tokens and $10 per million output tokens. A footnote in Google's post says that after Nov. 30, 2026, the price becomes $4 and $20.

$2 in, $10 out, until Nov. 30.

Google pitches Carbon as the model "built for tasks that test the limits of a model's intelligence," and Argon as the choice for most software engineering and knowledge work. Google also says Carbon uses more tokens than Argon.

The agent benchmarks are strong, and they are Google's

Google says Carbon ranks first on CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench, at less than half the cost per task of the next best outside model. The post gives rankings for those three, not raw scores.

On safety, Google says Carbon passed 92% of tasks on a 375-task agentic honesty benchmark, against 70% for Argon. It says the per-attempt breach rate in simulated cyberattacks fell to 0.8%, from 3.1% on Argon.

Every one of those numbers comes from Google. No independent lab has published its own yet.

Yes, another frontier model with another leaderboard sweep. The difference is the terms: Google is letting you test its top model at its mid-tier price, for three weeks, before you commit.

What to do before Dec. 1

Google says Carbon is for long-horizon agent workflows, cybersecurity and hard research. If your work looks like that, run your own evals on it now, on your real tasks.

Measure cost per finished task, not cost per token. A model that uses more tokens can cost more per job even at the same rate, and from Dec. 1 the rate doubles too.

If Argon already does the job, keep it. Google itself says Argon is the pick for most coding and knowledge work.

If your agent reads untrusted web pages or email, test for prompt injection yourself. Google says Carbon scored a 0% attack success rate on the Gray Swan IPI benchmark, and in the same post calls robustness "an area for continued improvement."

Watch the outside numbers and the calendar

The first thing to watch is independent testing. Leaderboard firms and your own logs will show whether the cost-per-task claim holds outside Google's tests.

The second is the date. Argon's introductory price also runs to Nov. 30, as TopFive reported in October, so every Gemini 4 bill you pay changes next month.

Test it this month. Decide with your invoice, not their chart.