TL;DR: Google DeepMind announced Gemini 4 Carbon on October 28, 2026, calling it the company’s strongest model for cybersecurity and long-horizon tasks. Google says it ranks first on version 5.0 of the Artificial Analysis Intelligence Index with a score of 59 and leads three security benchmarks. But you can’t use it yet: access starts with a small group of cyber defenders and expands through Google’s vetted Fairwind Program before any public release. Introductory API pricing is $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 later.
Gemini 4 Carbon arrives just two weeks after Gemini 4 Argon, and Google is pitching it less as a chatbot upgrade and more as a tool for finding and fixing software vulnerabilities. Here is what Google announced, what the benchmark claims mean, who can get access and what it will cost.
What Google announced
In the official announcement, Koray Kavukcuoglu, SVP at Google DeepMind and Google’s Chief AI Architect, described Carbon as “our strongest model with state-of-the-art cybersecurity and long horizon capabilities.” The key claims, all from Google itself, are:
- Top overall ranking: #1 on the recently updated v5.0 of the Artificial Analysis Intelligence Index, with a score of 59.
- Security benchmarks: 88.2% on SEC-bench Pro (finding vulnerabilities), 54.8% on ExploitGym (exploitation) and 71.0% on Collinear’s CWE-bench (patching).
- Web development: a jump of 89 Elo points to fourth place (1,768) on WebDev Arena.
- Long tasks: Google says its teams have seen Carbon “run for days” on multi-step engineering work such as root-causing and patching subtle bugs across large codebases.
Google also says the model was evaluated on the external security benchmarks “without guardrails,” which is worth noting: the safety filters a public version would ship with were not part of those scores.
Why Google is focusing on cybersecurity
Models that can find bugs can help attackers as easily as defenders. That tension shapes the whole launch. Google frames Carbon around the “cybersecurity defense lifecycle”: detecting a vulnerability, checking whether it can actually be exploited, then writing a fix at the root cause.
The company points to internal results to back the pitch. According to Google, Carbon more than doubles Argon’s success rate at generating secure code on internal benchmarks, and the Chrome team reported that it performed “more than 10X better than another lab’s large frontier model” at generating high-quality patches. Google did not name the other lab. Wiz, the cloud security company Google describes as its own platform, also reports Carbon leading on its web penetration testing benchmarks.
This is part of a wider trend in which AI labs are racing to put security-focused models in defenders’ hands first. We covered a similar move when Anthropic launched a “Cyber Mission” for critical infrastructure, and the risk on the other side is real: researchers have already documented a hacker using AI agents to breach more than 100 companies.
Safety and alignment claims
Google says Carbon uses the same multi-layered safety stack as Argon, combining training, real-time misuse defenses, prompt-injection resilience and chain-of-thought monitoring. The company lists several improvements, again based on its own testing:
| Area | Google’s claim for Carbon |
|---|---|
| Benchmark cheating | Successful cheating on Terminal-Bench 4.0 cut to 0.0%, with attempts down to 0.4% |
| Misalignment in live coding | High-severity misalignment alerts down almost 80% with internal teams |
| Honesty | About 2x fewer unsupported claims than Argon |
| Following instructions | Overreach down 26% and under-service down 29% |
| Prompt injection | 0.0% attack success on Gray Swan’s Indirect Prompt Injection benchmark, versus 0.7% for Argon |
These numbers have not been independently replicated yet, and “internal testing” results are hard for outsiders to check. Still, the focus on fabrication is notable given how often AI hallucination undermines agentic coding tools.
Who can use Gemini 4 Carbon?
For now, almost no one outside a small security circle. Google says that starting this week it will make Carbon available to “a subset of cyber defenders” and will “gradually phase access to additional trusted partners” through the Fairwind Program. The company also says it is taking part in the U.S. government’s voluntary process for pre-release model testing.
The Fairwind Program page describes a tightly controlled scheme aimed at high-priority defenders such as governments, healthcare providers and telecom services. According to Google, participants must:
- Pass background checks on their security history and ethical record.
- Use user-level authentication, phishing-resistant MFA and access controls.
- Limit access to internal security, incident response or penetration testing teams and track employee use.
- Not share, redistribute or sell access to the model.
Note that, at the time of writing, the Fairwind page still describes access to Gemini 4 Argon, not Carbon. Google has not given a date for general availability to developers, businesses or consumers, saying only that it wants to bring Carbon to those groups “as soon as possible.”
Gemini 4 Carbon pricing
When it does reach the API, Carbon will start at the same introductory price as Argon:
| Per 1 million tokens | Introductory price | Later price |
|---|---|---|
| Input | $2 | $4 |
| Output | $10 | $20 |
| Cached input | 95% off the input price | Not separately stated |
As VentureBeat points out, Google has not said when Carbon’s introductory period ends, while Argon’s higher rates are set to take effect December 1. If you are new to how this kind of billing works, our explainer on tokens in AI covers the basics.
What it means for developers and businesses
For most developers, Carbon is a preview of where Google’s public models are heading rather than something to plan around today. A few practical takeaways:
- Keep building on Argon. It is the model you can actually use now, and Carbon is designed as its successor at the same introductory price.
- Security teams should look at Fairwind. If your organization defends critical systems, the program is the only route to early access.
- Treat benchmark wins cautiously. Leaderboards shift quickly. OpenAI’s GPT-6.1 Terra landed just days ago, and independent testing of Carbon will matter more than launch-day charts.
The bottom line
Gemini 4 Carbon is Google’s attempt to show it can push frontier capability and safety forward at the same time, with a launch that puts cyber defenders ahead of everyone else. The benchmark and alignment numbers are impressive on paper, but they are Google’s own, and the model is not publicly available. Until Google sets a release date, Carbon is best understood as a signal: the next round of AI competition is about long-running agents and security, not just chat.
The Teck Hustlers Editorial Team covers technology, business, finance and gaming news for readers who want the story behind the headline. Every article is researched, written and edited by our in-house team, checked against company announcements, filings and reputable reporting, and updated when new facts emerge. We focus on what each development means for founders, professionals, investors and everyday users. Have a tip or a correction? Reach us through our contact page.
