Google announced Gemini 4 Carbon on October 28, 2026, a new frontier model it says sets state-of-the-art results in cybersecurity and long-horizon work. For now, only a subset of cyber defenders can use it.
That matters because Carbon lands about two weeks after Gemini 4 Argon, and its first users are the people who patch software the rest of us rely on. If you hoped to try it in the Gemini app this week, Google’s post does not say when that will happen.
TL;DR: Gemini 4 Carbon is Google’s strongest model, tuned for finding, validating and patching security flaws. Access starts with a subset of cyber defenders, then trusted partners, with no public date. Google’s benchmark numbers are self-reported, so treat them as a claim to test, not a settled ranking.
What Gemini 4 Carbon is built to do
Google’s post, written by Koray Kavukcuoglu, frames Carbon around the full defense cycle: detecting vulnerabilities, validating them, and fixing them. It also says the model is built for long-horizon work, with internal teams watching it run for days and persist through dead ends to find the root cause of subtle bugs.
On the Artificial Analysis Intelligence Index v5.0, Google says Carbon ranks first with a score of 59. The security results are where it leans hardest.
| Benchmark | What it tests | Carbon result |
|---|---|---|
| SEC-bench Pro | Finding real vulnerabilities | 88.2% |
| ExploitGym | Validating risk through exploitation | 54.8% |
| CWE-bench (Collinear) | Patching at the root cause | 71.0% |
| WebDev Arena | Web development | 1769 Elo, #4 |
Google adds that external benchmarks were run without guardrails, and that its Chrome team found Carbon more than 10 times better than another lab’s large frontier model at producing high-quality patches. Self-reported numbers are a starting point, not a verdict, and outside testing will say more once more people can run the model.
Who can use Gemini 4 Carbon right now
Almost no one. Google says access begins this week with a subset of cyber defenders, then widens gradually to more trusted partners through its Fairwind Program.
Developers, businesses and consumers get access “as soon as possible,” according to Google, after more feedback and guardrail work. The post gives no date, and it does not name the Gemini app, the API or Google Cloud as launch surfaces. Google is also taking part in the U.S. government’s voluntary pre-release model testing process.
That staged approach also explains why a cybersecurity model arrives behind a gate. A tool that finds and exploits flaws helps defenders and attackers alike, so the order of who gets it first is part of the product.
Pricing and the safety claims
Google lists introductory pricing of $2 per million input tokens and $10 per million output tokens, the same as Argon. Cached input is 95% cheaper, and the price rises to $4 and $20 per million tokens after the introductory period.
On safety, Google says Carbon uses the same layered approach as Argon, including real-time misuse defenses and chain-of-thought monitoring. It reports high-severity misalignment alerts in live internal coding environments dropped by almost 80%, and prompt-injection attack success on the Gray Swan benchmark fell from 0.7% to 0.0%.
A 0.0% result on one benchmark does not mean the problem is solved. It means that test no longer catches the model out, which is useful but narrower.
What to watch before it reaches regular users
For most people, Carbon’s first visible effect could be indirect: faster fixes in software like Chrome, whose team Google says tested the model. Chrome is already adding protections of its own, such as its suspicious site warnings.
The open questions are practical ones. When does the Gemini app get it, will the API price hold after the introductory period, and do independent tests land near Google’s own numbers? Until Google answers those, Carbon is mostly a security story rather than a product you can open on your phone.
