Perplexity Sonar Deep Research

Perplexity Sonar Deep Research belongs to a new generation of AI research agents that go beyond answering questions from training data to actively investigating topics on the web. Rather than relying on a fixed knowledge cutoff, this model autonomously breaks down complex queries into sub-tasks, searches and reads hundreds of pages, cross-references information, and refines its approach as it gathers findings. The result is a comprehensive, well-structured report that synthesizes evidence across domains such as finance, technology, health, and current events. Its 128,000-token context window supports sustained research sessions where maintaining coherence over long documents and many sources matters. The model distinguishes between standard prompt tokens, citation tokens from web searches, and a separate reasoning step where it works through gathered material before generating final outputs.

Perplexity built its deep research capability quickly after the category emerged in early 2025, positioning it as a fast, accessible alternative for developers and researchers who want programmatic access to a research agent. The model completes most research runs in under three minutes, and its pay-as-you-go pricing makes it practical for individuals and teams to integrate deep research into workflows without subscription lock-in. Pricing is tiered: input tokens cover both user prompts and processed citation tokens, search operations are billed per thousand queries, and a distinct reasoning token rate covers the model's analytical pass over gathered sources before drafting its response. This layered cost structure reflects how deep research is a multi-stage process rather than a single inference call. For users who need cited, exhaustive research delivered rapidly through an API, Sonar Deep Research targets a practical sweet spot between capability and accessibility.