enzosoler.comreviews 09bench 9 deviceslive test Can a ring and a watch replace my Apple Watch?
Published Updated

Gemini 4 Carbon is now on Google AI Ultra and the paid API: price and who can use it

A faceted iridescent crystal surrounded by floating white number 4s, Google's Gemini 4 Carbon artwork
Image: Google

Gemini 4 Carbon, Google's newest Gemini model, reached AI Ultra subscribers and the paid Gemini API on November 10. API price: $2 and $10 per million tokens through November 30, then $4 and $20.

Update, November 17: Carbon is no longer limited to security teams. In a November 10 post, Google said Carbon is on the paid Gemini API in AI Studio and in Gemini for business, and is rolling out to all Google AI Ultra subscribers in the Gemini app, AI Mode and Antigravity. AI Ultra starts at $99.99 a month. The API price is $2 per million input tokens and $10 per million output tokens through November 30, then $4 and $20, the same schedule as Argon. Google’s post doesn’t mention AI Plus, AI Pro or Workspace. Our AI plans comparison has every Google tier side by side. The original story from October 29 follows.

Update, November 12: Google released Gemini 4 Hydrogen, a faster companion model. In Antigravity, a new Orchestrator mode lets Carbon hand routine steps to Hydrogen. Full story: Gemini 4 Hydrogen: $1.50 on the API, now in Search and the Gemini app.

Gemini 4 Carbon is Google’s newest Gemini model, and for now only vetted security teams can use it. Google announced it on October 28, two weeks after Gemini 4 Argon reached paid subscribers. Google pitches Carbon as the stronger of the two.

According to Google’s blog post, access starts this week for a subset of cyber defenders. It expands to more trusted partners through Google’s Fairwind Program. Google says it will bring Carbon to developers, businesses and consumers “as soon as possible.” It names no date, and its post doesn’t mention the Gemini app, Workspace or any subscription plan. 9to5Google reported the launch the same day.

What Google says it can do

Everything below is Google’s own reporting. Independent tests are not out.

  • Ranking. Google says Carbon is first on the Artificial Analysis Intelligence Index v5.0 with a score of 59, and fourth on WebDev Arena at 1768 Elo, up 89 points.
  • Security work. The model scores 88.2% on SEC-bench Pro, which measures finding vulnerabilities. It reaches 54.8% on ExploitGym and 71.0% on CWE-bench, which measures patching. Google says these ran without guardrails.
  • Long jobs. Google says internal teams have run Carbon for days on multi-step engineering work, such as finding the root cause of a bug and patching it.
  • Behavior. Google reports about half as many unsupported claims as Argon, and almost 80% fewer high-severity misalignment alerts in live coding. Its prompt-injection test shows 0.0% attack success, against 0.7% for Argon.

What it costs

The API prices match Argon’s. Google lists $2 per million input tokens and $10 per million output tokens at launch, and $4 and $20 after the introductory period. Cached input costs 95% less. Google’s post doesn’t give the end date for the introductory price. For Argon, 9to5Google reported November 30.

Because only a small group has access, most teams can’t pay that price today. Nothing in Google’s post says Carbon will cost more than Argon in the consumer plans when it arrives. It doesn’t say it will cost the same either.

What it means if you pay for AI seats

Don’t change anything this week. Carbon isn’t available to you, and Argon is already in Google’s $19.99 AI Pro plan, as our Argon report covers.

Do two things. First, if you build on the Gemini API, run your evaluations on Argon now, while the introductory price holds. A second model at the same price tier means you can swap it in later and compare. Second, if your company runs on Google Workspace, ask your admin whether Argon and Carbon will reach your Workspace license. Personal AI plans sit outside your admin controls and data retention settings.

The security numbers matter if you run a small team without a security hire. A model that finds and patches bugs well helps attackers too, and Google restricted Carbon to defenders first. Google hasn’t said how long that restriction lasts. Treat the pace of patching on your own code and dependencies as a bigger risk than before.

Related, November 17: Gemini now splits chat history from AI training, and turning history off keeps chats for two months.

Related, November 17: OpenAI released GPT-6.2 Astra and Sol, a week after Claude Fable 5.2 and Gemini 4 Hydrogen.

Related, November 18: Google’s Genie 4 world model reaches AI Ultra subscribers.

One email when a review goes up. Unsubscribe from any of them.