Google announces Gemini 4 Carbon, with cyber defenders first

Expand

Open original image ↗

Google announced Gemini 4 Carbon on 28 October. Selected cyber defenders get first access; broader availability will follow. 1

The clearest independent result is already visible. Artificial Analysis lists Carbon (High) first on its Intelligence Index v5.0, at 59, followed by Claude Opus 5.5 at 58 in its Max configuration with the default fallback. The next Opus configuration scores 57. 2

A lead across a revised benchmark suite

The v5.0 index combines nine evaluations across knowledge work, coding, science and general capabilities. Each of those four groups contributes a quarter of the total. The ranking measures performance across that mix; individual tasks can favour a different model. 2

Google says internal teams have used Carbon for engineering tasks lasting days. Its reported cyber results include 88.2% on SEC-bench Pro, 54.8% on ExploitGym and 71.0% on CWE-bench, evaluated without guardrails. 1

Expand

Open original image ↗

Who gets access first?

The Fairwind Program prioritises governments, critical infrastructure and major technology platforms. Its stated aim is to give defenders time to find and fix vulnerabilities before comparable capabilities become more widely available. Applications go through a Cloud account team, with organisations vetted for security history and research practice. 3

Google plans to phase Carbon into that programme while collecting feedback. The announcement gives no date for general availability. 1

Introductory pricing

The announced launch rates are $2 per million input tokens and $10 per million output tokens, with cached input discounted by 95%. After the introductory period, input and output rise to $4 and $20. Google does not give an expiry date. 1

We covered the earlier Carbon references alongside Argon in our first report. There is now an official announcement and an independent index result to compare with those early signs. Our own creation-test scores remain a separate comparison.