Perplexity's multi-step research model that searches, reads and synthesizes sources into long, cited reports instead of single-pass answers.
Output Speed *
Intelligence Index *
Context Window *
Input price
Output price
Performance
Where Sonar Deep Research Spends Its Compute: Search, Reasoning, Synthesis
Technical Specifications
What the model supports
Limitations & Trade-offs
Best-Fit Workloads
Where this model earns its place
Market and competitive research reports
Perplexity positions the model for comprehensive report generation across domains like finance and technology. It searches, reads and evaluates sources, then synthesizes them. This suits analyst workflows that produce briefing documents, landscape reviews or due-diligence summaries. Because depth drives cost and duration, run it as a background or queued job rather than inside a latency-sensitive interface.
Current-events and topic monitoring digests
The model's live web retrieval makes it suitable for synthesizing recent developments on health, technology or news topics into a structured digest. Its multi-step search lets it cross-check several sources before writing. The trade-off is metered searching, so scheduled digests with well-scoped briefs are more economical than open-ended, frequent polling.
Research agents that escalate from simple queries
In a tiered architecture, a lightweight Sonar model can answer quick grounded questions. Sonar Deep Research handles briefs that need investigation. The 128K context fits long briefs and accumulated source material. This pattern keeps spend focused on the queries where multi-step retrieval and a separate reasoning phase change the answer.
Source-grounded drafting for knowledge teams
Teams drafting white papers, internal memos or literature-style overviews can use the model to gather and synthesize sources first. A human reviewer then checks the cited material. Quality is not independently benchmarked, so keep a verification step for claims and citations before publication.