Sonar

Sonar sits at the entry level of Perplexity's Sonar range of search-oriented models, positioned as a lightweight, cost-effective option for quick factual lookups rather than deep reasoning. Perplexity's documentation frames it as a non-reasoning model optimized for straightforward Q&A, topic summaries, product comparisons, and current-events queries where grounded information retrieval is enough, while steering users toward Sonar Pro or the reasoning tier for multi-step analysis. The model's defining characteristic is tight integration with Perplexity's real-time web search pipeline, so responses come with citations tied to fresh source material rather than relying solely on parametric knowledge.

For practical deployment, Sonar is shaped for speed and low cost: Perplexity prices it at one dollar per million input tokens and the same for output tokens, and it ships with a broad context window suited to digesting retrieved web content along with a user prompt. The documentation also signals a notable lifecycle change, noting that Sonar Chat Completions has been folded into Perplexity's Agent API and that the Sonar endpoint will be supported only through late September 2026, so teams building on it should plan around that migration path. The best fit is therefore high-volume, latency-sensitive applications such as in-product FAQ, lookup assistants, and summarization of recent material, where grounded citations and quick turnaround matter more than open-ended reasoning.