Gemini 4 Carbon is Google DeepMind's newest frontier model, announced on October 28, 2026, and Google calls it its strongest model so far. You almost certainly cannot use it yet. Access started that week with a small group of vetted cyber defenders, it will widen to more partners in Google's Fairwind Program, and Google gives no date for developers, businesses or the Gemini app. When it does reach the API, Google says it will cost $2 per million input tokens and $10 per million output tokens at first, then $4 and $20.
This explainer covers what Carbon is, how it differs from Gemini 4 Argon, who can get it, what it will cost, what Google claims about its safety, and what to use in the meantime. Facts are checked against Google's announcement, Google DeepMind's model pages, the Gemini API docs and the Artificial Analysis leaderboard as of the morning of October 30, 2026.
What is Gemini 4 Carbon?
Gemini 4 Carbon is the second model in Google's Gemini 4 family. Koray Kavukcuoglu, Google DeepMind's senior vice president and Google's chief AI architect, announced it in a Google blog post on October 28, 2026. Google describes it as built on Gemini 4 Argon, which reached the public two weeks earlier, and aims it at two jobs: long software engineering tasks and cyber defense.
Google says its internal teams have seen Carbon keep working for days on hard engineering problems, such as finding the root cause of subtle bugs across large codebases and patching them. The Verge reported the same claim and noted that Carbon tops some benchmarks against the best models from OpenAI and Anthropic.
The maker's profile is on our Google DeepMind page, and Carbon sits with this year's other launches on the AI model release timeline.
Gemini 4 Carbon release date and who can use it
There are two answers to "when is the Gemini 4 Carbon release date?" The announcement date is October 28, 2026. A public release date does not exist yet.
Google's post sets out the rollout in three steps:
| Stage | Who | When |
|---|---|---|
| 1 | A subset of cyber defenders | Starting the week of October 28, 2026 |
| 2 | More trusted partners through the Fairwind Program | Gradually, no dates |
| 3 | Developers, businesses and consumers on Google's products | "As soon as possible," no date |
Google says it is still gathering feedback and adjusting guardrails before step three, and that it is taking part in the US government's voluntary process for testing models before release. Reports from India Today and The Verge describe the same limited rollout.
When we checked on October 30, the Gemini API models page still listed only gemini-4-argon under Gemini 4. There is no Carbon model ID to call, and Carbon does not appear in Google AI Studio.
Argon gives a precedent, but not a promise. Google announced Argon on September 30 for Fairwind partners only and opened it to paid API customers and Google AI Pro and Ultra subscribers on October 13, as our Gemini 4 Argon release date guide explains. Google has not said Carbon will follow the same timetable, and its Carbon post does not say which group comes after Fairwind.
Gemini 4 Carbon price
Google has already set the API price, even though nobody outside the early cohort can buy it yet:
| Period | Input per million tokens | Output per million tokens | Cached input |
|---|---|---|---|
| Introductory | $2 | $10 | 95% off the input price |
| After the introductory period | $4 | $20 | Not stated |
That is the same introductory price Argon launched at, which WebProNews and The Verge also report. Google has not said when Carbon's introductory period ends. For Argon, Google's pricing page sets the end at November 30, 2026, with the higher price from December 1. Google also has not said which Gemini subscription plans will include Carbon. If you mainly want to compare subscriptions, see ChatGPT Plus vs Claude Pro vs Google AI Pro.
Gemini 4 Carbon vs Argon, Opus 5.5 and GPT-6 Astra
The headline number is the independent Artificial Analysis Intelligence Index, version 5. On October 30 its leaderboard put Gemini 4 Carbon (high) first at 59, ahead of Claude Opus 5.5 at 58 and GPT-6 Astra at 57. Gemini 4 Argon (high) scored 50 on the same version of the index.
Google DeepMind's Gemini 4 page adds a wider comparison table. A few rows show where Carbon leads and where it does not:
| Benchmark (Google's table) | Gemini 4 Carbon | Gemini 4 Argon | Claude Opus 5.5 | GPT-6 Astra |
|---|---|---|---|---|
| Artificial Analysis Index v5.0 | 59 | 50 | 58 | 57 |
| SEC-bench Pro (finding vulnerabilities) | 88.2% | 83.4% | not listed | 85.4% |
| ExploitGym (exploitation) | 54.8% | 42.0% | 34.5% | 42.4% |
| CWE-bench (patching) | 71.0% | 68.0% | 67.0% | 68.0% |
| FrontierSWE 2 (agentic coding) | 61.2% | 55.0% | 62.3% | 65.5% |
| OSWorld-2.1 (computer use) | 81.5% | 74.7% | 87.2% | 80.9% |
Two things to keep in mind. First, Google says its outside cybersecurity results were measured without guardrails, so customers using the guarded public model may not see the same scores. Second, Google's methodology notes say most rival numbers come from those labs' own reports, and several Carbon scores are computed by Google itself. Benchmark methods are a topic of their own; here the point is simply that Carbon leads the general index and the cyber tests while Opus 5.5 and GPT-6 Astra still win some coding and computer use rows.
For which Claude model to pick today, see Claude Fable vs Opus vs Mythos, and the Claude Opus 5.5 profile has its specs.
Why Google is holding Carbon back
The reason is the same skill that makes Carbon useful. A model that is very good at finding and exploiting software flaws helps defenders patch systems, and it could also help attackers. Google's answer is to give defenders a head start.
The Fairwind Program is how that works. Its page says Google works with more than 650 partners and gives priority to defenders such as governments, healthcare providers and telecommunications services. Google runs background checks on applying organizations. Partners must use phishing resistant multifactor authentication, may give access only to their internal security, incident response or penetration testing teams, and may not share or resell access. On October 30 the page still described partner access to Argon rather than Carbon.
This staged approach is similar to how other labs gate their most capable models. Our guide to Preparedness Framework vs Responsible Scaling Policy compares the OpenAI and Anthropic rules behind those decisions.
What Google claims about Carbon's safety
Google says Carbon uses the same layered safety stack as Argon (training, live misuse defenses, prompt injection resistance and monitoring of the model's chain of thought) and that it pushed further on alignment. Its main claims, all from Google's own testing:
- Fewer shortcuts: successful cheating on Terminal-Bench 4.0 fell to 0.0%, with attempts at 0.4%, while accuracy improved.
- Fewer alarms: high severity misalignment alerts in live coding work with Google's own teams dropped by almost 80%.
- More honest output: about half as many unsupported claims as Argon, according to Google and The Verge's summary.
- Prompt injection: a 0.0% attack success rate on Gray Swan's indirect prompt injection benchmark, against 0.7% for Argon.
These are vendor claims, not independent audits. They matter because the safety story this autumn has been about agents that act on their own, as in the OpenAI sandbox escape. For the bigger picture, read what AI safety is.
Pros and cons of Gemini 4 Carbon
Pros
- First place on the Artificial Analysis Intelligence Index v5 as of October 30, 2026.
- Strong cyber defense results across finding, exploiting and patching flaws (Google reported).
- Built for long tasks that run for hours or days.
- Launch price matches Argon's introductory rate, and Google has published the later price too.
- Google reports fewer shortcuts, fewer unsupported claims and stronger prompt injection resistance than Argon.
Cons
- No public access, no model ID and no release date for developers or consumers.
- Fairwind admits organizations, not individuals, and only for internal security teams.
- Cyber scores were measured without guardrails, and many figures are self reported.
- Rivals still lead some coding and computer use tests in Google's own table.
- No word yet on which Gemini plans will include it or when the introductory price ends.
What to use while you wait
If you need the best Google model today, use Gemini 4 Argon. It is in the Gemini app for Google AI Pro and Ultra subscribers and on the paid Gemini API as gemini-4-argon. Our Gemini 4 Argon profile lists its context window and prices.
If you are open to other labs, Claude with Opus 5.5 and ChatGPT with GPT-6 Astra sit just behind Carbon on the Artificial Analysis index and are available now on paid plans. For a side by side look at the four main assistants, start with ChatGPT vs Claude vs Gemini vs Grok.
If you work on security for a government, critical infrastructure operator or large technology platform, the Fairwind page is where to apply.
FAQ
Can I use Gemini 4 Carbon now?
Not unless your organization is in Google's first cyber defender cohort or the Fairwind Program. As of October 30, 2026, Carbon is not in the Gemini app, Google AI Studio or the public Gemini API.
When is the Gemini 4 Carbon release date for everyone?
Google has not given one. It announced Carbon on October 28, 2026 and says developers, businesses and consumers will get it "as soon as possible" after more feedback and guardrail work.
How much will Gemini 4 Carbon cost?
At launch, $2 per million input tokens and $10 per million output tokens, with cached input 95% off. After the introductory period the price rises to $4 and $20. Google has not said when that period ends.
Is Gemini 4 Carbon better than Claude Opus 5.5?
On the Artificial Analysis Intelligence Index v5 it leads by one point, 59 to 58, and it leads Google's cyber tests. Opus 5.5 still scores higher on some coding and computer use tests in Google's own table, and you can use Opus 5.5 today.
What is the difference between Gemini 4 Carbon and Argon?
Argon is the Gemini 4 model you can use now, generally available since October 13, 2026. Carbon is built on Argon, scores higher on Google's tests, especially in cybersecurity and long tasks, and is still limited to vetted defenders.
What is the Fairwind Program?
Google DeepMind's early access program for cyber defenders. Vetted organizations such as governments, healthcare providers and telecommunications services get advanced models first, under strict security terms and only for defensive work.
