Sonar Deep Research

Sonar Deep Research is built as a research-oriented reasoning model that autonomously searches and evaluates large sets of web sources before synthesizing its findings into a single, expert-style report. According to Perplexity's official documentation, it is designed to perform exhaustive searches across hundreds of sources and produce detailed analyses rather than short answers, positioning it for workflows that demand comprehensive coverage of a topic. The model exposes a dedicated reasoning token track, alongside separate charges for citations and search queries, which reflects how its long-form research behavior is broken into distinct retrieval and synthesis stages behind the API.

In practice, the model fits use cases such as academic literature reviews, market and technology briefings, health information digests, and other analytical tasks where stitching together many references matters more than a quick reply. Reasoning and temperature controls are exposed to developers, and pricing is structured around per-million-token rates for input, output, reasoning, and citation tokens, with an additional charge per batch of search queries, making cost predictable for long research sessions. The broader Sonar line is migrating toward Perplexity's Agent API with continued support through late September 2026, so teams building on Sonar Deep Research today should plan a migration path while taking advantage of its strong fit for deep, citation-heavy investigations.