Under the hood, Perplexity runs on OpenAI’s GPT-3.5 model along with its own AI model — a variant of Meta’s open-source Llama 2 model. Users who upgrade to the Pro version can choose between a handful of different models, including GPT-4 and Anthropic’s Claude. (I used GPT-4 for most of my searches, but I didn’t see much of a difference in the quality of the answers when I chose other models.)
Perplexity is also refreshingly good at admitting when it doesn’t know something. Sometimes, it gave a partial response to my question, with a caveat like “No further details are provided in the search results.” Most AI chat products I’ve used lack this kind of humility; their responses sound confident even when they’re spouting nonsense.