Sonar (Perplexity API)

Sonar is the API layer that Perplexity exposes for developers who want to integrate Perplexity's search-and-answer engine into their own products. Where calling a standard LLM API gives you a model that knows only what it was trained on (up to a cutoff date), calling Sonar gives you a model that searches the live web, synthesizes results, and returns an answer with citations. The underlying model is based on Meta's Llama, fine-tuned for search and retrieval tasks.

The API is structured similarly to OpenAI's Chat Completions format, making it straightforward to slot into existing codebases that already call an LLM API. Pricing is usage-based per query. In February 2026, Perplexity also released companion embedding models (pplx-embed-v1 and pplx-embed-context-v1) under the MIT license, positioned as best-in-class for dense semantic retrieval tasks.

For builders, Sonar is relevant for any application where users need current information rather than information frozen at a training cutoff: news, financial data, recent events, product pricing, and other time-sensitive queries. It is a useful alternative or complement to building your own RAG pipeline with a vector database, especially for general-knowledge queries that are hard to anticipate and pre-index.