Perplexity CEO calls Gemini 4 Carbon the strongest research model; early WANDR scores lead
Post cites Perplexity's CEO and early WANDR results: Gemini 4 Carbon scores 0.857 at $5.86 per task, ahead of Argon's 0.802 and GPT-6 Astra's 0.735, with a planned Perplexity Computer rollout.
Google may have just raised the bar for AI research models. Perplexity CEO Aravind Srinivas says Gemini 4 Carbon is the strongest model yet for long-form research and answer quality. The early WANDR benchmark results make that claim worth watching: → Gemini 4 Carbon: 0.857 score, $5.86 per task → Gemini 4 Argon: 0.802 → GPT-6 Astra: 0.735 Carbon leads on both answer quality and score-to-cost efficiency in the chart. Perplexity also plans to bring it to the Computer feature for Pro and Max users, pending token capacity. The bigger takeaway: AI competition is shifting beyond raw intelligence. The winners must deliver stronger research, reliable answers, and better economics. The next model race won't just be about who scores highest. It will be about who delivers the most useful intelligence per dollar.
This post quotes another post: “Perplexity early evals: Gemini 4 Carbon tops long-form research”