What Is Gemini 4 Carbon?

Google announced Gemini 4 Carbon on October 28, 2026, a frontier model for cybersecurity and long-horizon work. Most people cannot use it yet. This guide covers what the announcement says about access, benchmarks, and pricing.

Daniel Carter·October 30, 2026

Explore AI Models

TL;DR:

Gemini 4 Carbon is Google's latest frontier AI model, announced on October 28, 2026. It is built for cybersecurity and long-horizon tasks, and it is not generally available yet. Access starts with selected cyber defenders and trusted partners in the Fairwind Program.

Gemini 4 Carbon: Overview and Release Date

Gemini 4 Carbon is Google DeepMind's latest frontier model for cybersecurity and long-horizon tasks. Google announced it on October 28, 2026. The announcement was authored by Koray Kavukcuoglu, SVP of Google DeepMind and Chief AI Architect at Google.

Building on Gemini 4 Argon, Google's earlier frontier model, Carbon extends capabilities in coding, agentic workflows, and cybersecurity. Its key advancements focus on security research and complex engineering tasks that require sustained work over extended periods.

Who Can Use Gemini 4 Carbon Right Now?

Gemini 4 Carbon is not yet generally available. Google is first rolling it out to selected cyber defenders, followed by additional trusted partners through its Fairwind Program.

Who	Status in the October 28, 2026 announcement
Selected cyber defenders	First group. Access starts this week for a subset of defenders.
Trusted partners	Further access through the Fairwind Program, expanded over time.
Developers, businesses, and consumers	Not generally available. Google says it will open access as soon as possible, with no date.

The announcement does not describe a public way to use Gemini 4 Carbon in Google AI Studio or the Gemini API, and it does not give a public application form for general users.

Gemini 4 Carbon vs Gemini 4 Argon

Carbon follows Argon in the same model line. Argon remains the earlier frontier model for complex reasoning and real-world workflows, including work that already covers cybersecurity. Carbon's announcement puts more weight on cybersecurity and long-horizon engineering.

Comparison	Gemini 4 Argon	Gemini 4 Carbon
Primary focus	Complex reasoning and real-world workflows	Cybersecurity and long-horizon engineering
Coding	Complex software engineering tasks	Further coverage of longer coding tasks
Agentic workflows	Multi-step autonomous tasks	Complex engineering tasks that can continue for days
Secure code generation	Baseline in Google's comparison	More than 2x uplift in success rate for generating secure code
Unsupported claims	Baseline in Google's comparison	Cuts unsupported claims by approximately 2x
Prompt injection resistance	0.7% attack success rate on the evaluated Gray Swan IPI benchmark	0.0% attack success rate on the same evaluated benchmark

Carbon is an evolution of Argon rather than a completely different product category, with Google's reported improvements concentrated in coding, cybersecurity, and alignment.

The secure-code result is a success-rate uplift for generating secure code. The unsupported-claims result means approximately 2x fewer unsupported claims in Google's internal comparison.

For the earlier model, see Google's Gemini 4 Argon announcement.

Coding and Long-Horizon Performance

Google describes Carbon as able to stay on large engineering jobs for days, including debugging and patching inside large codebases. That description comes from how internal teams are using the model.

Result	Figure from the announcement	What it covers
Long-horizon coding	Work that can continue for days	Internal use on large-codebase debugging and patches.
WebDev Arena	4th place, 1768 Elo, up 89 Elo	One web-development evaluation. It is not an overall coding rank.
Artificial Analysis Intelligence Index v5.0	1st place, score of 59	This version of the index. Scores from earlier versions may not be directly comparable.

Carbon scored 59 on the Artificial Analysis Intelligence Index v5.0, ranking first in this version of the evaluation. Scores from different index versions may not be directly comparable. The WebDev Arena rank and Elo are as reported on October 28, 2026, and can change if the board updates.

Gemini 4 Carbon Cybersecurity Benchmarks

Google reports Carbon's results across three established external cybersecurity benchmarks. The results cover three stages of the cybersecurity defense lifecycle: finding vulnerabilities, validating exploitability, and patching vulnerabilities at their root cause.

Stage	Benchmark	Result
Finding real vulnerabilities	SEC-bench Pro	88.2%
Validating exploitability	ExploitGym	54.8%
Patching vulnerabilities at the root cause	CWE-bench	71.0%

Source: Google DeepMind, October 28, 2026. These external cybersecurity benchmarks were evaluated without guardrails.

ExploitGym tests whether vulnerabilities can be exploited, matching Google's description of validating risk through exploitation.

Google also reports two observations alongside those benchmarks.

Google's Chrome team reported that Carbon performed more than 10 times better than another lab's large frontier model at generating high-quality patches. This is an internal team report from Google, not an independent public benchmark.

Wiz observed that Carbon led on a web penetration-testing benchmark. Google's announcement does not include a score for that test.

Gemini 4 Carbon Safety and Alignment

Gemini 4 Carbon builds on Argon's multi-layered safety and security framework, with further improvements in alignment, prompt injection resistance, and reliable task execution.

Topic	Reported figure	How to read it
Preventing cheating	Terminal-Bench 4.0: 0.0% successful cheating, 0.4% cheating attempts	Fewer cases of finishing a task by taking a shortcut, on this benchmark.
Improving alignment	High-severity misalignment alerts down by nearly 80%	An internal observation, not an external leaderboard.
Better task execution	Overreach down 26%, under-service down 29%	On multi-step plans, the model stayed inside the requested scope more often in Google's measurement.
Resisting prompt injection	Gray Swan IPI: 0.0% for Carbon, 0.7% for Argon	A 0.0% result reflects performance on this specific benchmark, not a guarantee of complete security.

Gemini 4 Carbon Pricing

Google's Carbon announcement lists introductory pricing and the rate that applies after the introductory period. General access is not open yet, so these are announced prices rather than a confirmed public rate.

Price type	Input	Output
Introductory pricing, as announced	$2 per 1 million tokens	$10 per 1 million tokens
After the introductory period	$4 per 1 million tokens	$20 per 1 million tokens

Cached input tokens are priced at 95% off the applicable input token rate.

Google has not specified how long the introductory pricing will last.

If you need product photos, marketing visuals, or social media content today, you can continue using dedicated AI image and design tools rather than waiting for Carbon's broader availability. On Designkit, explore the AI Product Photo Generator for product visuals or the AI Image Generator for marketing and social media graphics.

Conclusion

Gemini 4 Carbon is Google's new frontier model for cybersecurity and long-horizon engineering. For now, access starts with selected cyber defenders and trusted partners in the Fairwind Program. The open question is when developers, businesses, and consumers get a published entry point, and what product surfaces Google attaches to the model after that.

Primary source: Introducing Gemini 4 Carbon.

Frequently Asked Questions

What is Gemini 4 Carbon?

Gemini 4 Carbon is Google DeepMind's frontier model announced on October 28, 2026. It is built for cybersecurity and long-horizon tasks, and it extends Gemini 4 Argon in coding, agentic workflows, and security research.

Is Gemini 4 Carbon available to the public?

No. Gemini 4 Carbon is not generally available. Google is first rolling it out to selected cyber defenders, then to additional trusted partners through the Fairwind Program. Developers, businesses, and consumers do not have general access yet.

How is Gemini 4 Carbon different from Gemini 4 Argon?

Gemini 4 Carbon builds on Argon with reported improvements in coding, cybersecurity, and alignment. Google reports more than a 2x uplift in secure code generation success and approximately 2x fewer unsupported claims in internal testing.

What are Gemini 4 Carbon's cybersecurity benchmarks?

In Google's October 28, 2026 announcement, SEC-bench Pro is 88.2% for finding real vulnerabilities, ExploitGym is 54.8% for validating exploitability, and CWE-bench is 71.0% for patching vulnerabilities at the root cause. These benchmarks were evaluated without guardrails.

How much does Gemini 4 Carbon cost?

Google's announcement lists introductory pricing of $2 per million input tokens and $10 per million output tokens. After the introductory period, the stated rates are $4 and $20 per million tokens. Google has not said when that period ends. Cached input tokens are priced at 95% off the applicable input token rate.

When can developers use Gemini 4 Carbon?

Google says it will bring Gemini 4 Carbon to developers, businesses, and consumers as soon as possible. The announcement does not give a date or a public signup. It also does not say that Carbon is already available in Google AI Studio or the Gemini API.

Can Gemini 4 Carbon generate ecommerce product photos?

Google's announcement focuses on cybersecurity, coding, and long-horizon tasks. It does not introduce ecommerce product photo generation as a supported capability. If you need product photos or marketing visuals now, use a dedicated image tool instead of waiting on this model.