# Google releases Gemini 4 Carbon first to cyber defenders

> Google's new frontier model scores 59 on the Artificial Analysis index and costs $2 per million input tokens, but only trusted security teams can use it for now.

URL: https://claudegpt.si/news/gemini-4-carbon-cyber-defenders  
Published: 2026-10-28T18:50:00.792Z  
Author: ClaudeGPT Si Newsroom  
Section: AI Models  
Topics: Google, Gemini, Google DeepMind, Language models, Cybersecurity  
Publisher: ClaudeGPT Si

## Key takeaways

- Google announced Gemini 4 Carbon on October 28, with access limited at first to a subset of cyber defenders through the Fairwind Program.
- Google lists an introductory price of $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 later.
- Google reports a score of 59 on the Artificial Analysis Intelligence Index v5.0, ranked first, and 88.2% on SEC-bench Pro.

Google announced Gemini 4 Carbon on October 28, a new frontier model it calls its "strongest model with state-of-the-art cybersecurity and long horizon capabilities," in a [post by Koray Kavukcuoglu](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/). You can't use it yet unless you are a cyber defender: access starts with a small group and widens through Google's Fairwind Program.

That makes Carbon the second major Gemini 4 model in about two weeks, after [Gemini 4 Argon](https://claudegpt.si/news/is-gemini-4-argon-free). It also changes who gets the strongest Gemini first. Google is giving it to people who defend software before it goes to everyone else.

## Who can use Gemini 4 Carbon

Starting this week, Carbon is available to a subset of cyber defenders. Google says access will grow gradually to other trusted partners through the Fairwind Program. Developers, businesses and consumers get it "as soon as possible," but Google gives no date and no word on which apps, APIs or subscription plans will carry it. It says it is still collecting feedback and tuning its guardrails, and that it is taking part in the U.S. government's voluntary pre-release testing of models.

## Price: the same as Argon for now

Google lists an introductory price of $2 per million input tokens and $10 per million output tokens, the same as Argon. Cached input tokens cost 95% less. After the introductory period the price rises to $4 per million input tokens and $20 per million output tokens. The post does not say when that period ends, and it does not state a context window.

## What Google says it can do

All the numbers below come from Google's post, and Google says it ran the security benchmarks without guardrails.

| Benchmark | Carbon | Note |
| --- | --- | --- |
| Artificial Analysis Intelligence Index v5.0 | 59 | Ranked first |
| SEC-bench Pro (finding vulnerabilities) | 88.2% | Google calls it state of the art |
| ExploitGym (exploiting them) | 54.8% | Google calls it a new state of the art |
| CWE-bench (patching) | 71.0% | |
| WebDev Arena | 1768 Elo | Fourth place, up 89 points |
| Gray Swan indirect prompt injection | 0.0% attack success | Argon scored 0.7% |

On Terminal-Bench 4.0, a test of agents working in a terminal, Google says successful cheating fell to 0.0%, with cheating attempts at 0.4%.

Google also reports results from its own teams. It says Carbon more than doubled Argon's success rate at writing secure code, and that the Chrome team found it more than 10 times better than another lab's large frontier model at producing high-quality patches. It says unsupported claims fell by about half against Argon, and that high-severity misalignment alerts in live internal coding environments dropped by almost 80%. None of those internal results has been checked by outsiders.

## How the ranking compares

The index score is the one number you can check. Our report on [Artificial Analysis Intelligence Index v5](https://claudegpt.si/news/artificial-analysis-intelligence-index-v5) had Claude Opus 5.5 leading at 58, so Carbon's 59 puts it one point ahead on that index. A one-point gap is small, and the figure is the one Google chose to cite.

## What we don't know yet

Google has not published an API model page, a context window, a release date for general access or the list of plans that will include Carbon. The post links a safety and security report, which is where to look for how Google tested the model before wider release.

Watch for the general rollout. Until then, the price and the benchmark scores are the only parts of this launch that anyone outside the defender group can act on.

## Sources

- [Gemini 4 Carbon: our next frontier model](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/), Google