Sonar

Sonar is the core model line that powers Perplexity’s platform, designed around tight integration with retrieval. Unlike generic LLMs, Sonar is tuned to ground its answers in evidence, producing well-structured and citation-rich outputs that reduce hallucinations. It builds on the R1 reasoning backbone, which emphasizes chain-of-thought reasoning, step-by-step analysis, and higher accuracy on complex benchmarks like math, coding, and multi-hop Q&A.

The Sonar lineup includes smaller, faster models for real-time assistants and customer-facing copilots, as well as larger ones such as Sonar Large 32K for document-level reasoning and research-style tasks. Sonar Pro sits in the middle as the dependable workhorse, balancing accuracy with latency. All variants support structured JSON outputs, long-context inputs, and tool/function calling, making them easy to deploy in retrieval pipelines and enterprise agents.

In practice, Sonar powers Perplexity’s own app and API, ensuring that when users ask questions, the model not only reasons through them but also cites sources so responses can be verified. Enterprises and developers also adopt Sonar for search copilots, knowledge assistants, and analytics explainers, where grounded answers and predictable quality are more important than flashy generative capabilities.