Skip to content

AI

Google's Gemini 4 Carbon is out, and it wants your hardest work

Google DeepMind calls Carbon its strongest model yet, with frontier-level cybersecurity and long-horizon agent skills. Everyone from developers to consumers can try it today.

Cover photo: Maurizio Pesce from Milan, Italia / CC BY 2.0

TL;DR

  1. Google made Gemini 4 Carbon available to developers, enterprises, scientists and consumers on Nov 10, 2026.
  2. Google calls it its strongest model, with frontier cybersecurity and long-horizon agentic workflows.
  3. Argon stays the pick for most work; Carbon targets tasks that test a model's limits.

What happened

On Nov 10, 2026, Google announced that Gemini 4 Carbon is available to developers, enterprises, scientists and consumers. Google describes it as its strongest model, with frontier performance in cybersecurity and in long-horizon agentic workflows. Those span complex knowledge work, software engineering and scientific reasoning.

The rollout has a back story. Two weeks earlier, Google released Carbon to a set of cybersecurity defenders through its Fairwind Program. So the defenders got it first, and now everyone else gets a turn.

Carbon also arrives right after the launch of Gemini 4 Argon. Google says Carbon pushes the frontier even further than Argon does. Argon, it says, remains the ideal choice for most software engineering and knowledge work.

Carbon is for the hard stuff: tasks that test the limits of a model's intelligence, such as complex, multi-step coding and debugging.

Why it matters

This is a frontier release from a major lab, and it lands in front of every audience at once. No waitlist for consumers, no enterprise-only gate. The cybersecurity framing is the notable part: Google is claiming frontier-level performance there and gave defenders early access before the general launch.

The two-model split is the other signal. Google is telling you Argon covers the majority of work and Carbon is the heavy equipment. That is a quiet admission that most of your tasks do not need the strongest model in the building.

Mind the sourcing. The capability claims come from Google's own announcement, and the excerpt we have carries no benchmark numbers. "Frontier" is a vendor word until independent tests say otherwise.

Argon for the daily grind, Carbon for the tasks that test a model's limits.

What it means for Europe

Google presents the release as available to developers, enterprises, scientists and consumers, and the pitch is aimed at all of them. For European teams, that makes this a practical question first: can you reach it, and from where. Check your region and plan before you plan a migration.

The cybersecurity angle is worth a look for EU security teams, since defenders were the first group Google served. Procurement and compliance reviews will still apply, and they are your problem, not the model's.

What to do with it

  • Pick the right tier.Start tasks on Argon, and escalate to Carbon only when multi-step coding or debugging stalls.
  • Test on your ugliest ticket.Feed it the bug your team has dodged for weeks and judge the result, not the keynote.
  • Ask security to look.If you run a defense team, find out how the Fairwind Program access translates to your plan.
  • Wait for independent numbers.Treat "frontier" as a claim until third-party evaluations show up.

Ciao, sapiens. Carbon is the new smartest thing in the room, and you still have the coffee budget.