Is Gemini 4 Carbon really the best model, or just the best marketing?

Number one by a single point, still behind on coding, and almost nobody can use it.

On 28 October Google showed off Gemini 4 Carbon, its new top-tier model. Artificial Analysis scored it straight away: 59 points on version 5 of its Intelligence Index. That puts it first, one point ahead of Claude Opus 5.5. It leads on hard reasoning, knowledge and science tasks, it hallucinates less, and a task costs about 23% less than with Opus.

Sounds like a crown. Look closer.

On agentic coding, the thing a huge chunk of paying users actually care about, it still loses to Opus, 59 to 62. The index itself was reshuffled into version 5 only a week earlier. And the lead is one point. One.

And then the funny part: you cannot have it. Access is limited to selected cyber-defence teams. Everyone else gets "as soon as possible", no date.

There is a real reason given, the model is very strong at cybersecurity. But notice the effect. A closed circle of chosen users creates an aura. If only the special people get it, it must be special. That is marketing gold, whether or not it was planned.

So what is this really? Mostly Google planting a flag. It has been a while since Google sat on top of a big leaderboard, and now it can say it does. On paper that is true. In practice, a one-point gap is noise, coders are better off where they are, and the model is locked away.

Not nothing. Just not the earthquake the headline suggests.