Part of our AI models explained: ChatGPT, Claude, Gemini and the rest guide
Google announced Gemini 4 Carbon on Oct 28, two weeks after Gemini 4 Argon, and called it "our strongest model." Most people can't use it yet. Google says it will make Carbon available "to a subset of cyber defenders" starting this week, then gradually add "trusted partners" through its Fairwind Program. Developers, businesses and consumers come later, "as soon as possible," with no date given.
The reason is what Carbon is good at. Google pitches it on cybersecurity and long, multi-step engineering work, and the same skill that finds and fixes a flaw can also be used to attack one. The company says it's taking part in the U.S. government's voluntary pre-release testing, and Sundar Pichai said on Alphabet's earnings call the same day that Carbon is "being tested by the U.S. government."
All of the performance figures so far come from Google. It says Carbon:
- ranks first on version 5.0 of the Artificial Analysis Intelligence Index, with a score of 59;
- scores 88.2% on SEC-bench Pro (finding vulnerabilities), 54.8% on ExploitGym (exploiting them) and 71.0% on Collinear's CWE-bench (patching them), measured without guardrails;
- gains 89 Elo points on WebDev Arena to rank fourth, at 1768;
- more than doubles Argon's success rate at writing secure code, on Google's internal tests.
Google also reports safety numbers. Carbon cut successful cheating on Terminal-Bench 4.0 to 0.0%, according to the company, and no attack succeeded in Gray Swan's indirect prompt injection benchmark, compared with 0.7% for Argon. Google says Carbon makes about half as many unsupported claims as Argon in its internal testing. None of this has been checked independently yet.
The first users are the kind of organizations the Fairwind Program already serves: governments, critical infrastructure operators in healthcare, telecoms and energy, and core technology platforms. Google says the program has more than 650 partners. Applicants are vetted, they must use phishing-resistant multi-factor authentication, and only internal security, incident response or penetration testing teams may get access. Until now, Fairwind partners have had exclusive access to a cyber-defense version of Argon.
The price is already set. Google says Carbon will launch at the same introductory price as Argon: $2 per million input tokens and $10 per million output tokens, with cached input 95% off. After the introductory period, it rises to $4 and $20. Google didn't say when that period ends for Carbon. Argon's ends on Nov 30, as we covered in Gemini 4 Argon for builders.
For builders, Argon remains the Gemini model you can call today. To see how it compares on cost with Claude and other models, see our LLM API pricing comparison. If you're giving any model access to your code or tools, the same rules still apply: run coding agents with tight permissions.
Sources
- Google, “Gemini 4 Carbon: our next frontier model” (Koray Kavukcuoglu, Oct 28, 2026)blog.google
- Google DeepMind, Fairwind Programdeepmind.google
- Google, “Q3 2026 earnings call: Remarks from our CEO” (Oct 28, 2026)blog.google
- Google, “Gemini 4 Argon: Built to solve your hardest problems” (Oct 13, 2026)blog.google
Topics
John F.AI desk
John F. is a Codema.ai house byline: a pen name used by our editorial desk, not one individual journalist. It covers the AI beat: new models and how they're priced, open models, AI coding tools and agents, and security incidents that affect AI developers. Pieces under this name are written with AI assistance and reviewed by a human editor before they're published.
All stories by John F.


