Google's Gemini 4 Carbon Tops AI Rankings, Nipping Claude Opus 5.5 by a Point

Google has a new frontier model, and it's sitting at the top of the pile. Gemini 4 Carbon took the number one spot on Artificial Analysis' Intelligence Index V5, edging out Claude Opus 5.5 by a single point, according to Tom's Hardware. The model landed with more noise than most releases, and the benchmark sheet backs up some of the hype.

The margin is razor thin at the very top, but Carbon's spread is wider than that one point suggests. It claims the best score on four of the nine evaluations in the suite. It also runs cheaper per task than other models near the summit, including Claude Opus 5.5 — a detail that tends to matter more to buyers than a single index point.

The sharper weapon might be accuracy. Carbon tops Artificial Analysis' new AA-Omniscience v1.1 test, which measures hallucinations alongside how well a model recalls facts. That's the number Google will want plastered on a slide. Governments and institutions weighing AI adoption care about trust and security as much as raw capability, and owning the lowest hallucination score is a straightforward pitch.

Google isn't handing the keys to everyone at once. Access is going to cyber defenders first through its Fairwind program, with broader availability coming at some unspecified point after that. It's a familiar playbook: get the model into the hands of people defending networks, gather credibility, then open the doors wider.

For now, the frontier race comes down to a point, a price tag and a hallucination score. Artificial Analysis' rankings are one scoreboard, not the final word, but they're the scoreboard rivals watch. Google just moved to the front of it.