Google DeepMind launches Gemini 4 Carbon frontier model

Google announced Gemini 4 Carbon on October 28, calling it its strongest model yet, with a particular focus on cybersecurity defense and long, multi-step engineering tasks. Google says Carbon ranks number 1 on the newly updated Artificial Analysis Intelligence Index v5 with a score of 59, and on cybersecurity benchmarks it hit 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation testing, and 71.0% on CWE-bench for patching root causes. Artificial Analysis separately confirmed Carbon as the new number 1 on its index, and said it ranks second only to Claude Opus 5.5 on the Coding Agent Index.

Google says internal teams have run Carbon for days straight on complex engineering problems, and that it cut high-severity misalignment alerts by almost 80% and attempted cheating on Terminal-Bench 4.0 down to 0.4%, while fully resisting a prompt-injection benchmark where its predecessor, Argon, already scored well.

Carbon is not generally available yet. Google is rolling it out gradually, starting this week with a small group of trusted cyber defenders, then expanding to partners through its Fairwind Program before wider release to developers, businesses and consumers. Google says pricing will start at Argon's introductory rate of $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 after the introductory period.

Why it matters

A model explicitly tuned for finding, exploiting and patching vulnerabilities raises the stakes for both defenders and attackers racing to use AI first.

Source: Google