Perplexity’s Deep Research now lives inside Computer, a cloud system that orchestrates up to 20 frontier models for each query. If you spend hours manually gathering sources, cross‑checking facts, and building reports, the new workflow removes that grind. The model automatically breaks a complex question into subtasks, routes each to the best‑suited model, and runs thousands of retrieval steps in parallel using Search as Code. This means you get a cited, ready‑to‑use report, deck, or dashboard without stitching together separate tools.
Common pain points and how Deep Research in Computer solves them:
Inconsistent accuracy – Legacy deep research relied on a fixed pipeline, missing nuanced details. The code‑driven approach adapts retrieval on the fly, raising BrowseComp scores from 40.7% to 83.8% and improving performance on expert benchmarks.
Scattered data sources – You often need internal files (PDFs, spreadsheets) plus live web data. Computer reads your uploaded files alongside the web, cites every claim inline, and cross‑references with premium sources like PitchBook and CB Insights.
Manual model selection – Choosing the right AI for legal reasoning, data analysis, or writing is time‑consuming. The system automatically assigns subtasks to specialized models (e.g., a legal model for contract review, a data model for variance checks) while using Opus 4.6 as the core reasoning engine.
Deliverable friction – Turning raw research into a presentation or spreadsheet usually requires extra steps. Deep Research outputs work‑ready reports, briefs, decks, dashboards, and live spreadsheets that you can preview, approve, or reject directly inside Computer.
Cost uncertainty – The feature is available to Perplexity Max subscribers; developers can access the same Agentic Search stack via the pay‑as‑you‑go Agent API, letting you scale usage to match budget.
Verification need – Although every claim is cited, outputs still benefit from a quick human review. Treat the generated report as a first draft, verify critical numbers, and then finalize.
By letting the model handle the heavy lifting of search, synthesis, and formatting, you free up time for analysis and decision‑making. Use the preset deep‑research endpoint, upload your internal documents, and let Computer deliver a cited, actionable output.