Google announces Gemini 4 Carbon, starting with cyber defenders only

Monday, 2 November 2026By Sridhar Mukkandi1 min read

Google announced Gemini 4 Carbon on 28 October, two weeks after Gemini 4 Argon, calling it its strongest model, with state-of-the-art cybersecurity and long-horizon capabilities. It goes first to a subset of cyber defenders, with access extended to more trusted partners through Google's Fairwind Program. Google says broader availability will follow once it has gathered feedback and refined guardrails, and that it is taking part in the U.S. government's voluntary pre-release model testing.

Google reports 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym and 71.0% on CWE-bench for patching, and notes those cybersecurity benchmarks were run without guardrails. It also reports first place on the Artificial Analysis Intelligence Index v5.0, with a score of 59. Introductory pricing is $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 afterwards.

So what

A model that is better at finding vulnerabilities is better at exploiting them too, which is why access is staged. The figures are Google's own: test any new model on your own tasks before trusting a leaderboard.

SourceGoogleblog.google/innovation-and-ai/models-and-research/gemini-models/ Go deeper · EssayYour LLM's 0.95 isn't a probability

Drafted with AI from the original source, then checked and edited by Sridhar Mukkandi. Spotted a mistake? Write to hello@tarkika.com and we'll correct it here.

More from The Signal

Agents GitHub: agents started more than 40% of pull requests in SeptemberA merge rate measures what reviewers accepted, not what held up in production. If agents now write the first draft, track reverts and follow-up fixes on their changes, and spend review time where a mistake costs most. GitHub Blog Agents IBM makes Agent Identity generally available: an owner and short-lived access for every agentYou don't need a product to start: a list of every agent, its owner, the tools it may call and when its access ends is most of the value. Any agent without an owner should be read-only. IBM Research Microsoft’s ThinkingBox turns database checks into rewards for training agentsChecking the state an agent leaves behind is a more honest test than reading its transcript, and the same check doubles as a training reward. Start with the three actions your agent takes most often. Microsoft and Hugging FaceAll stories