Gemini 4 Carbon tops WANDR research eval

Gemini 4 Carbon’s most consequential test may be whether it can turn benchmark strength into dependable, sustained research inside an agentic workflow.

• –WANDR is closely aligned with Perplexity’s research product, making this result highly relevant but not fully independent
• –Perplexity’s planned Computer rollout could give developers an early way to test Carbon on real multi-step research tasks
• –Google’s initial access remains limited, so third-party availability may matter more than the headline benchmark
• –Independent evaluations will need to confirm whether Carbon’s answer quality holds across domains, sources, and longer sessions
• –Stronger research models could shift competition from one-shot chat quality toward research reliability, citation accuracy, and task completion