Sonar

Sonar is an AI model built by Perplexity that, instead of just guessing an answer from memory, searches the internet, pulls up real web pages, and builds its answer based on what it finds there. A regular language model answers purely from what it learned during training, but Sonar works more like a research assistant who pulls a few reference sources from the library, reads them, and summarizes the findings.

The names sonar and sonar-pro refer to different versions within the same family, differing in speed and answer quality. Other companies and developers who want to add this search capability to their own products can call Sonar through intermediary services like OpenRouter.

The catch is that which web pages Sonar chooses as its sources directly determines how trustworthy its answer is. If it relies on an obscure site with almost no visitors, users end up treating weak information as reliable without realizing it.

The article explains that it "accessed Perplexity's two models, sonar and sonar-pro, through OpenRouter... and asked each of the 380 purchase-intent categories." It's worth keeping straight that Sonar is the model made by Perplexity, while OpenRouter is merely the intermediary channel used to call it.

Ask Perplexity about a specific software category — for example, type "recommend the best project management software." Then click through the source links attached to the answer one by one and check whether those sites are actually well-known or get much traffic. It's good practice for judging for yourself how reliable an answer's grounding really is.