Sonar by Perplexity

Sonar is Perplexity AI’s specialized suite of models designed to provide fast, web-grounded answers with real-time citations. Built on a refined Llama 3.3 backbone, Sonar models are optimized for developers who need to integrate high-speed search and retrieval-augmented generation (RAG) into their applications. Unlike traditional LLMs that rely on static training data, Sonar retrieves live internet data at query time to ensure factual accuracy and provides inline source citations. The pipeline supports various search depths (Low, Medium, High) to balance latency and thoroughness, and includes features like domain filtering, recency filters, and academic search modes.