Gemini 4 Carbon: Google's new #1 AI model, coming soon

Short answer: On October 28, 2026, Google announced Gemini 4 Carbon, its new frontier AI model, just two weeks after Gemini 4 Argon. Google says it ranks #1 on the Artificial Analysis Intelligence Index v5.0 and makes about half as many unsupported claims as Argon in internal tests. For now only a small group of cybersecurity defenders can use it; Google says developers, businesses and consumers get it "as soon as possible".

What happened

Google DeepMind's Koray Kavukcuoglu introduced Carbon as Google's "strongest model with state-of-the-art cybersecurity and long horizon capabilities". According to Google, it scores 59 on version 5.0 of the Artificial Analysis Intelligence Index, the top spot, and The Verge reports that it sits ahead of Claude Opus 5.5 and GPT-6 Astra on that ranking. Google also says Carbon climbed 89 Elo points to #4 on WebDev Arena.

The headline use is security work. Google says Carbon leads benchmarks for finding real software vulnerabilities (88.2% on SEC-bench Pro) and fixing them at the root cause (71.0% on CWE-bench), and that it generated secure code at more than twice the success rate of Argon in internal tests. Internally, teams saw it work on hard engineering tasks for days. All of these are Google's own figures, not independent tests.

Access is deliberately slow. Carbon goes first to a subset of cyber defenders this week, then to more trusted partners through Google's Fairwind Program, while Google keeps testing guardrails and takes part in the U.S. government's voluntary pre-release testing.

Why people are talking about it

Two frontier models in about 15 days is a fast pace, and the launch dominated tech feeds. On Hacker News, the announcement reached the #2 spot of the day with more than 880 points and over 700 comments, and The Verge, 9to5Google and VentureBeat covered it on launch day.

What it means if you use AI

• More honest answers are the goal. Google says Carbon cuts unsupported claims by about 2x versus Argon and adds caveats instead of making things up. If it holds up in real use, that is good news for anyone who relies on a chatbot for facts.
• It asks before guessing. Google says that when a request is unclear, Carbon checks sources and points out discrepancies before acting.
• Better resistance to hidden instructions. Google reports a 0.0% attack success rate on Gray Swan's indirect prompt injection benchmark, which matters as assistants start reading your email and web pages for you.
• Not in your app yet. There is no date for the Gemini app, and features and regions can differ at launch.

Quick checklist

1. Don't look for Carbon in the Gemini app today; Google has not given a consumer date.
2. Treat "#1" as one benchmark ranking, not a promise about every task.
3. When it arrives, test it on questions you already know the answer to.
4. Keep private details out of prompts in any AI tool, whatever the model.

FAQ

Can I use Gemini 4 Carbon now? Only if you are one of the selected cyber defenders. Wider access is planned after more guardrail testing.

Facts come from Google's announcement of October 28, 2026, with reporting from The Verge and 9to5Google. Benchmarks are reported by Google and Artificial Analysis.

Related reading:

• Genie 4 preview: AI worlds with sound you steer by text
• ChatGPT vs Claude vs Gemini for everyday life
• Why AI makes things up: hallucinations explained

Cover image: Demis Hassabis, CEO of Google DeepMind, at the 2024 Nobel Lectures in Stockholm. Photo: Jay Dixit, Wikimedia Commons, CC BY-SA 4.0.