Perplexity (Sonar)

Sonar is a search-grounded model rather than a conventional one: it retrieves from the live web and answers from what it finds, with citations, rather than from training data alone.

That makes it a different category of tool. The question is not how it compares on reasoning benchmarks, but whether your task needs current information with sources attached.

Currency and attribution together. For questions where the answer changed recently — pricing, product features, current events, who holds a role — a search-grounded model is structurally more reliable than a large model working from a training cutoff.

Citations make the output checkable. For research work, being able to follow a claim to its source is often more valuable than the claim itself, because it turns the model into a starting point rather than an authority.

Answer quality is bounded by search quality. If the web is wrong, thin, or dominated by SEO content on a topic, Sonar reflects that — it is a retrieval system with a language model attached, and retrieval is the weaker link.

Reasoning depth is not its strength. Complex analysis, code and creative work are all better served by a conventional flagship.

Use Sonar for research and current-information tasks where citations matter, and treat its output as sourced leads to verify rather than settled answers. Pair it with a reasoning model — Sonar to gather, something stronger to analyse — rather than expecting one model to do both.