Perplexity Deep Research

Research agent that searches, reads and writes cited reports on complex questions.

Free; Pro $20 per monthAutonomous web research agent

No user reviews yet. Be the first to rate it.

Visit website
Perplexity Deep Research website screenshot
Perplexity Deep Research website

Quick Facts

DeveloperPerplexity
Websiteperplexity.ai
Agent typeAutonomous web research agent
Best forAnalysts, students, consultants and marketers who need a sourced overview of a topic quickly.
Not ideal forWork that depends on paywalled or internal data the agent cannot reach, or anyone who will not check citations.
Pricing modelFreemium subscription; per-token API
Starting priceFree; Pro $20 per month
Pricing detailsFree: limited use. Pro: $20 a month or $200 a year, about 20 Research queries a day. Max: $200 a month, unlimited. Enterprise Pro: $40 per seat a month, 500 a day. Enterprise Max: $325 per seat a month. Education Pro: $10 a month. API (Sonar Deep Research): $2 per million input and $8 per million output tokens plus search fees.
Free planLimited
Free trialFree plan available
What it does on its ownPlans and runs many search rounds, reads sources and runs code on its own, then writes a report; users then ask follow-ups.
PlatformsWeb, iOS, Android, Mac, Windows, Comet browser, API
Agent launched2025
Details last verified28 October 2026
CategoriesFree AI Agents, Personal Assistants & Productivity, Research
User ratingNo reviews yet

Perplexity’s Deep Research gives free users a taste and Pro subscribers about 20 full research runs a day for $20 a month. Ask it a complex question and it runs search after search, reads hundreds of sources, and within minutes writes a report with citations for each claim.

Perplexity launched the feature in February 2025, when it scored 21.1% on the Humanity’s Last Exam benchmark. In the app it sits among the search modes, and Perplexity’s plan pages now count usage as Research queries.

How It Works

After a user picks the research mode and types a question, the agent plans a series of searches, reads what it finds, refines its queries and repeats. It can run code in a sandbox for calculations. The finished report streams into an editable document, and since a 2026 upgrade it can also produce structured outputs such as presentations, spreadsheets, dashboards and simple websites. The model is chosen automatically in this mode; paid users have been moved onto Anthropic’s Claude Opus models.

Deep Research works on the web and in Perplexity’s apps for iOS, Android, Mac and Windows, and inside its Comet browser. Developers can call the same capability through the Sonar Deep Research API.

Pricing

The free plan includes limited research use. Pro costs $20 a month or $200 a year and allows about 20 Research queries a day, while Max costs $200 a month with no cap. For organizations, Enterprise Pro runs $40 per seat a month with 500 queries a day, and Enterprise Max costs $325 per seat. Verified students and educators can get Pro for $10 a month. On the API, Sonar Deep Research costs $2 per million input tokens and $8 per million output tokens, plus charges for citations, reasoning and searches, so one full query can cost $0.41 or more.

What to Watch Out For

Limits shift: at launch in 2025 Pro users were offered 500 research queries a day, against about 20 now, so check your plan’s current allowance. When it runs out, queries reportedly fall back to standard search. The agent reads the open web, so paywalled journals and internal documents are out of reach unless uploaded. Citations make checking easy, and checking is still necessary: a confident summary can misread a source.

Features

  • Many rounds of web search on a single question
  • Reads hundreds of sources per report
  • Inline citations for every claim
  • Runs code in a sandbox for calculations
  • Structured outputs such as presentations, spreadsheets, dashboards and websites
  • Runs on Claude Opus models for paid users
  • API access through Sonar Deep Research

Pros and Cons

Pros

  • Usable for free, with a low-cost Pro plan
  • Fast compared with manual research, usually minutes
  • Every report is cited, so claims can be checked
  • Available on web, mobile, desktop and through an API

Cons

  • Research allowances vary by plan and can shift
  • Cannot choose the model manually in this mode
  • Quality depends on what is openly available on the web
  • Long reports still need fact-checking against the cited sources
Write a Review