On 28 October 2026, Google announced Gemini 4 Carbon, which it describes as its strongest model, with state-of-the-art cybersecurity and long-horizon coding abilities. It arrived two weeks after Gemini 4 Argon. Carbon is not available to developers or consumers yet. Google is releasing it first to a subset of cyber defenders and will expand access gradually through its Fairwind Program.
What Carbon is built for
Google positions Carbon for two kinds of work. The first is cyber defence, which covers finding vulnerabilities in software, checking whether they can be exploited and writing patches. The second is long-running coding work, and Google says its own teams have seen Carbon run for days on a single job. All the figures below come from Google's announcement, and we have not seen independent testing yet.
The numbers Google is reporting
- 88.2% on SEC-bench Pro, a benchmark for finding vulnerabilities.
- 54.8% on ExploitGym, a benchmark for exploiting them.
- 71.0% on Collinear's CWE-bench, a benchmark for patching them.
- First place on version 5.0 of the Artificial Analysis Intelligence Index, with a score of 59.
Google says the security benchmarks were measured without guardrails switched on. It also says the Chrome team found Carbon produced high-quality patches more than ten times better than another lab's large frontier model, though the other lab is not named.
Why defenders get it first
A model that is good at finding security holes can help fix them or help someone break in. Google's answer is to put defenders at the front of the queue so they can patch first. The Fairwind Program, run by Google DeepMind, prioritises governments and national cyber authorities, operators of critical infrastructure such as healthcare, telecoms, energy and finance, and core technology platforms, and it welcomes academic labs working on defensive benchmarks. Google says it runs background checks on organisations that apply, and partners may only give access to internal security, incident response or penetration testing teams. This mirrors a long habit in security research, where people who find a bug tell the people who can fix it before telling the public.
Carbon also builds on the safety system Google used for Argon, which combines model training, real-time misuse defences, resistance to prompt injection and monitoring of the model's reasoning. Google says it is taking part in the US government's voluntary process for testing models before release.
When everyone else gets it, and what it will cost
Google says it is still gathering feedback and improving guardrails before making Carbon available to developers, businesses and consumers as soon as possible. It has not given a date. When it opens up, the introductory price will be $2 per million input tokens and $10 per million output tokens, matching Argon. After the introductory period, the price rises to $4 and $20.
What to do this week
Most businesses will feel Carbon through faster security fixes from the software they already use, so the simplest step is to make sure those fixes reach you. Switch on automatic updates for browsers, operating systems and phones across your team. Ask your IT or security provider how they use AI to find and fix vulnerabilities. If you work at a hospital, utility, bank or similar organisation, ask your Google Cloud account team whether Fairwind applies to you. We think defenders-first is the right order for a model this capable.
To see where Gemini fits among the main assistants today, visit the Gemini page in Learn, or read our explainer on Gemini's agent for work.