Perplexity released Deep Research inside its Computer platform on June 11, combining iterative multi-model search with computer-side task execution so that research and output production happen in a single workflow rather than across two separate sessions.
The change addresses a workflow friction the company says its data confirms matters: research and analysis account for 26% of Computer tasks, the single largest category ahead of coding and asset creation, per Perplexity. Previously, users had to run a Deep Research query in one thread, then open a separate Computer thread to turn findings into a document, slide deck, or dashboard. The integration removes that handoff: a single query now routes through 20 or more frontier models, breaks into parallel subtasks, and produces work-ready output in one pass.
The technical piece behind the integration is what Perplexity calls Search as Code. Instead of issuing one query at a time, Computer writes and executes code that designs a custom search program for each question, running potentially thousands of retrieval steps in parallel, evaluating source quality mid-run, and adjusting the search plan when early results fall short. The code runs in a secure sandbox alongside the model, so the model can see arriving results and change course before synthesis. Perplexity says the Perplexity Computer architecture scores higher on Humanity's Last Exam, BrowseComp, and DeepSearchQA compared to the prior separate-tool version, with improvements across factual accuracy, depth of analysis, and citation quality.
On the source side, Deep Research in Computer can pull from authorized internal connectors including files, apps, and connected data sources, alongside the live web. Premium data sources such as Statista, PitchBook, and CB Insights are automatically invoked when the query warrants it. Every claim in the Perplexity output carries an inline citation linked to a live source URL.
The practical implication for enterprise teams is that a complex multi-part research task runs through a single Computer session rather than requiring a separate research step followed by a drafting step. Perplexity's example in the announcement traces a buy-side diligence brief through parallel analyses of four separate jurisdictions, each producing a risk assessment reconciled into a single memo. How much of the benchmark improvement is attributable to the Search as Code architecture versus the simpler workflow consolidation will become clearer as third-party evaluations accumulate.
