Arena: Gemini 4 Carbon gets more cost-competitive as Agent Arena tasks get harder

Arena says its recent Agent Arena analysis found that Google DeepMind's Gemini 4 Carbon becomes more cost-competitive as tasks get more demanding. It says Carbon's savings relative to GPT-6 Astra and Claude Opus 5.5 grow on the most expensive sessions.

These are Arena's findings from its own users' sessions; the post gives no figures.