1300 166 138 Tell us what you need
News / Google

What is Gemini 4 Carbon, and when can Australians use it?

By Monika Tantau • Published 3 weeks ago
A laptop on a desk showing the Gemini logo with the words developed by DeepMind, as Google announces its Gemini 4 Carbon model

Key takeaways

• Google announced Gemini 4 Carbon on 28 October, calling it its strongest model so far
• Google says it ranks first on the Artificial Analysis Intelligence Index, with a score of 59
• For now only a group of cyber defenders can use it, with no public date yet
• The introductory API price works out at about $2.84 per million input tokens

Most Read

Google has announced Gemini 4 Carbon, which it calls its strongest model so far, with leading results in cybersecurity and in long tasks that run for hours or days. Google says it ranks first on the Artificial Analysis Intelligence Index, according to its announcement.

The catch is access. Carbon is starting with a small group of cyber defenders, and Google has not said when everyone else gets it. Here is what it does, what it will cost Australian businesses and how long the wait might be.

What Gemini 4 Carbon is built for

Carbon arrives two weeks after Gemini 4 Argon and pushes further on security and long jobs. Google says its own teams have watched it run for days on complex engineering work, working through dead ends to find and patch subtle bugs across large codebases and low level systems.

On the Artificial Analysis Intelligence Index, version 5.0, Google reports a score of 59 and the top spot. That index changed version recently, so its scores are not directly comparable with older readings, and the rest of the results below are Google’s own.

Security results Google is highlighting

Google quotes 88.2% on SEC bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploiting them and 71% on CWE bench for patching them. It also says Carbon more than doubles Argon’s success rate at writing secure code, and that Chrome engineers found it more than ten times better at generating patches than another lab’s large model, which it does not name.

On safety, Google reports a 0% attack success rate on the Gray Swan prompt injection benchmark, down from 0.7% for Argon, and almost 80% fewer serious misalignment alerts in live coding environments. It uses the same layered protections as Argon, with a separate safety report.

When Australians can use Gemini 4 Carbon

For now, Carbon is limited to a subset of cyber defenders, then to trusted partners through Google’s Fairwind Program. Google says it will reach developers, businesses and consumers soon, without a date, and the announcement does not mention the Gemini app or any country limits.

The best guide is Argon. It also started with cyber defenders, and about two weeks later it was open to paying Gemini users in Australia. If Google follows the same path, Carbon could reach Google AI Pro and Ultra subscribers within weeks, but that is our estimate, not Google’s promise.

A hand holds a phone showing a large number 4 and the Gemini 4 Argon name, the earlier model whose launch price Gemini 4 Carbon matches

What Gemini 4 Carbon will cost

Google has already set the developer price. At the introductory rate, Carbon costs about $2.84 per million input tokens and $14.22 per million output tokens, rising to about $5.69 and $28.43 once the introduction ends. Cached input is 95% cheaper. Google bills in US dollars, so these move with the exchange rate.

Those are the same list prices Argon launched with, so Carbon does not add a premium per token. What a task actually costs will depend on how many tokens it uses, which is where our Argon vs Opus 5.5 comparison found the real differences.

Why Australian businesses should care

Carbon is aimed at security teams and long engineering jobs, which makes it relevant to anyone patching software or reviewing code for vulnerabilities. With the Australian Signals Directorate reporting a steady rise in cyber incidents, a model built to find and fix flaws faster is worth watching.

For most businesses the practical step is simple: keep using what works today and plan to test Carbon on real tasks when it opens up. Choosing between Gemini, Claude and ChatGPT models on price per task rather than headline scores belongs in your AI strategy.

Google’s lead may not last long

Top spots on these indexes change hands every few weeks, and every figure here comes from Google until independent tests arrive. The next signals to watch are a public release date and fresh scores from Artificial Analysis at different effort settings.

Scroll to Top