Gemini 4 Carbon: Google's cyber defense model

Q: How does Gemini 4 Carbon compare to Gemini 4 Argon?

Carbon is the second Gemini 4 model, two weeks after Argon. It scores 59 on the Artificial Analysis Intelligence Index v5.0 against Argon's 50, and Google reports more than twice Argon's success rate on generating secure code. Pricing is the same at both tiers.

Gemini 4 Carbon is Google's newest frontier model, announced on October 28, 2026, and built for cybersecurity and long engineering jobs. For now only a subset of vetted cyber defenders can use it. This page checks whether that includes you, states the price at both tiers, and separates Google's claims from what outside testing has confirmed.

Last verified: October 30, 2026 · Primary sources: Google DeepMind announcement coverage via Reuters and VentureBeat · Independent explainer, not affiliated with Google

Check if you can use it

AA index 59 (first) $2/$10 intro, then $4/$20 Access: gated Announced: Oct 28, 2026

Can you use Gemini 4 Carbon today?

Pick the description closest to you. The checker follows Google's rollout order, not a waitlist promise.

Who are you?

Google says access starts with a subset of cyber defenders this week, then widens to more trusted partners through Fairwind, and reaches developers, businesses, and consumers as soon as possible. No date is published for any later group, and the Fairwind page still listed Argon as its model on October 29, 2026.

What will Gemini 4 Carbon cost you?

Carbon launches at the same two-tier pricing as Argon. The calculator shows both tiers side by side, because the introductory end date is unpublished and most developers will arrive after it ends.

Monthly input (millions of tokens)

Monthly output (millions of tokens)

Cached input share (%)

Compare against

Budget at the standard $4/$20 rate. Google has not published when introductory pricing ends, and the public Gemini API docs still list Argon, not Carbon, as an available model, so the $2/$10 rate is a stated future price for most readers, not a price you can pay today.

What is Gemini 4 Carbon?

Carbon is the second model of the Gemini 4 generation, two weeks after Argon. Google calls it its strongest model, with state-of-the-art cybersecurity and long-horizon capabilities. In practice that means three jobs: find a flaw in software, prove it can be exploited, and write the fix. Google says its internal teams have run Carbon for days on single tasks, working through dead ends to root-cause subtle bugs in large codebases and low-level systems.

Beyond security, Google reports gains in coding, scientific tasks, full-stack web development, and multimodal understanding. On WebDev Arena, a public ranking of models that build websites, Carbon gained 89 Elo points to reach fourth place at 1769. That is a real, checkable ranking, and it sits below the security scores in importance: Carbon is sold on defense work first.

What Google has not published

Item	Status on October 30, 2026
Parameter count	n/a
Context window	n/a
Open weights	No, closed model only
General availability date	n/a
Introductory pricing end date	n/a
Public API model ID	n/a

Argon's context window (1 million tokens) does not transfer to Carbon. Any page that quotes a Carbon context window or parameter count on October 30 is guessing, unless it cites a newer Google source.

Benchmarks: what the scores actually say

Independent: Artificial Analysis Intelligence Index v5.0

Artificial Analysis places Carbon first on its Intelligence Index v5.0 with a score of 59, ahead of Claude Opus 5.5 at 58, GPT-6 Astra at 57, and Argon at 50. The lead over Opus 5.5 is one point on a composite index, not a win on every task. Independent coverage also notes Carbon trails Opus 5.5 on agentic coding and that its hallucination rate rose from Argon's 4% to 11% on one factual-knowledge test, still among the lowest measured. Treat the one-point lead as a narrow top rank, not a runaway.

Google's security numbers (vendor-reported, run without guardrails)

Benchmark	What it tests	Carbon
SEC-bench Pro	Finding real vulnerabilities	88.2%
ExploitGym	Proving a flaw can be exploited	54.8%
CWE-bench	Patching at the root cause	71.0%
WebDev Arena	Building websites (public ranking)	1769 Elo, 4th

Google says these external security evaluations ran without guardrails, so the numbers show raw capability with safety filters off. That is also why access is gated: the same capability that finds flaws for defenders can find them for anyone.

Google's internal claims (not independently verified)

More than twice Argon's success rate on generating secure code.
Chrome team result: more than 10 times better than another lab's large frontier model at high-quality patches. Google does not name the rival model.
High-severity misalignment alerts down almost 80% in internal live coding work, and unsupported claims down about half against Argon.
0.0% successful cheating on Terminal-Bench 4.0, and 0.0% attack success on Gray Swan's indirect prompt-injection benchmark (Argon: 0.7%).

Sources: Reuters, VentureBeat, Technology.org. Benchmark rows marked vendor-reported come from Google's announcement as relayed by these outlets.

Where you can use it

Surface	Who gets it
Fairwind Program, first subset	Selected cyber defenders, from the week of October 28, 2026
Fairwind Program, wider	More trusted partners, gradually, no date published
Gemini API	Not listed as an available model in public docs on October 30, 2026
Gemini app (Pro / Ultra)	No date published
Free tier	No, and no date published

Google says it is taking part in the US government's voluntary pre-release model testing, and will gather feedback and iterate on guardrails before wider release. Argon is the closest template: it went to Fairwind defenders on September 30, 2026 and reached app subscribers 13 days later. Carbon may follow a similar pace or a slower one, because its gating reason is offensive capability, not general readiness. Do not plan a launch around the Argon timeline.

How we got here

September 30, 2026: Argon reaches Fairwind defenders first, two weeks before general availability.
October 13, 2026: Gemini 4 Argon rolls out worldwide at $2/$10 introductory pricing.
October 28, 2026: Google DeepMind announces Gemini 4 Carbon, its strongest model, first to a subset of cyber defenders.
October 29, 2026: Reuters, VentureBeat, and other outlets report the launch, the $2/$10 price, and the gated rollout. The Fairwind page still lists Argon.
Later, date unpublished: More trusted partners, then developers, businesses, and consumers.

Frequently asked questions

Can I use Gemini 4 Carbon today?

Probably not. As of October 30, 2026, Google gives Carbon first to a subset of cyber defenders through Fairwind, then to more trusted partners. Developers, businesses, and Gemini app users have no published date yet. Use the checker above for your situation.

How much does Gemini 4 Carbon cost?

At introductory pricing: $2 per million input tokens and $10 per million output tokens, with cached input at 95% off ($0.10 per million). Standard pricing is $4 and $20. Google has not said when the introductory period ends, so budget at $4/$20.

How does Gemini 4 Carbon compare to Gemini 4 Argon?

Carbon scores 59 on the Artificial Analysis Intelligence Index v5.0 against Argon's 50, and Google reports more than twice Argon's success rate on generating secure code. Pricing is identical at both tiers. Argon is the model you can actually call today.

Is Gemini 4 Carbon open weights?

No. Carbon is a closed Google model, available only through the gated rollout. Google has not announced open weights, a public API model ID, or a self-hosted option.

What is the Fairwind Program?

Google's vetted-access program for cyber defenders. Fairwind members get early access to frontier models for defense work, under Google's conditions. Carbon starts with a subset of those defenders, then expands to more trusted partners.

What has Google not published about Carbon?

As of October 30, 2026: parameter count, context window, general availability date, the end date of introductory pricing, and a public API model ID. Treat any page that states those figures as guessing.

Why is Carbon gated when Argon was not?

Google ties the gating to capability: Carbon's headline scores are in finding and exploiting software flaws, so Google is testing safeguards and taking part in the US government's voluntary pre-release model testing before wider release.

Should I budget at the introductory or the standard price?

Budget at the standard $4/$20 rate. The introductory $2/$10 rate has no published end date, and by the time most developers get access it may already have ended.

Sources

Reuters · Google launches Gemini 4 Carbon AI model with cybersecurity focus (Oct 28, 2026)
VentureBeat · Google unveils Gemini 4 Carbon, retaking the world's no. 1 spot (Oct 28, 2026)
Technology.org · Google launches Gemini 4 Carbon for cyber defense (Oct 29, 2026)
MarkTechPost · Google DeepMind introduces Gemini 4 Carbon (Oct 28, 2026)
Deccan Herald · Gemini 4 Carbon: strongest AI model with cybersecurity focus (Oct 29, 2026)

Page published 2026-10-30. Prices and availability change; items marked n/a are unpublished, not zero.

More model guides

Agents

Machine-readable summary: gemini-4-carbon/llms.txt: pricing, benchmarks, access, links. Updated 2026-10-30.

Ask an AI about this page:

ChatGPT · Claude · Perplexity · Grok · Gemini