Google's New Gemini Model Tops an Independent AI Ranking, but Only Cyber Defenders Can Use It
Google released Gemini 4 Carbon on Oct. 28, 2026, and the model went to the top of an independent ranking the same week. Artificial Analysis, which scores commercial AI models on its own tests, places Carbon first on version 5.0 of its Intelligence Index, at 59 points. Almost nobody can use it yet: Google said access opens that week to a subset of cyber defenders.
Google says Carbon is built for engineering jobs that run for hours or days, and for security work: finding flaws in software, checking which of them can really be exploited and writing the patch. The company said it will widen access through its Fairwind Program and bring the model to developers, businesses and consumers after more work on its guardrails.
Artificial Analysis's public leaderboard ranked 77 models when it was read on Oct. 29, 2026, and the entry below Carbon is Claude Opus 5.5 at 58. The index scores a model in one named configuration rather than as a whole, and the score belongs to Carbon's high setting.
The cybersecurity numbers are Google's own. On three outside benchmarks, which it says it ran with the model's guardrails switched off, the company reports 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym and 71.0% on Collinear's CWE-bench for patching. No independent evaluation of those three results has been published.
Carbon ships at the same introductory price as Gemini 4 Argon, the model Google released two weeks earlier: $2 per million input tokens and $10 per million output tokens. Google also said it is taking part in a U.S. government voluntary process for pre-release model testing.
