Gemini 4 Carbon: Google's new top AI model leads on cybersecurity, priced from $2 (about Rs 554) per million tokens

Google has announced Gemini 4 Carbon, which it calls its strongest model yet, with "state-of-the-art cybersecurity and long horizon capabilities". It was unveiled on 28 October by Koray Kavukcuoglu, Google DeepMind's senior vice president and Google's chief AI architect, in a post on Google's blog, two weeks after Gemini 4 Argon (our story). The Express Tribune and The Verge also covered the launch.

Who can use it, and when

Not most people yet. This week access is limited to a subset of cyber defenders, widening gradually to trusted partners through what Google calls the Fairwind Program. Carbon is also going through the US government's voluntary pre-release testing. Google says it will bring the model to developers, businesses and consumers "as soon as possible" but gives no date, and does not say when it reaches the Gemini app or which plans will get it.

The price, in rupees

Cached input is 95% off. Rupee figures use about Rs 276.77 to the dollar; convert other amounts with our currency converter. Pakistani developers paying in dollars by card should add the bank's foreign transaction charges.

What Google says it can do

• No. 1 on the Artificial Analysis Intelligence Index v5.0, with a score of 59.
• 88.2% on SEC-bench Pro for finding software vulnerabilities and 54.8% on ExploitGym, both best results according to Google; 71.0% on CWE-bench for patching.
• More than twice Argon's success rate at writing secure code in Google's internal tests, and a 0.0% success rate for attackers on the Gray Swan indirect prompt-injection test, against 0.7% for Argon.
• Fourth on WebDev Arena, up 89 Elo points to 1768.
• Google's teams have run it for days on long engineering jobs such as tracing and patching bugs in large codebases.

Google also reports that cheating on the Terminal-Bench 4.0 test fell to 0.0% and high-severity misalignment alerts in live coding dropped by almost 80% compared with Argon. These are Google's own figures; independent tests will follow once more people have access.

Why the cautious rollout

A model that is very good at finding and exploiting software flaws is useful to defenders and attackers alike, which is why Google is giving it to security teams first. Google said in September that Gemini had broken into three real companies during an authorised test in May (our story).