Perplexity has added its Deep Research feature to Computer, extending the company’s agent-focused product with a tool designed to break down complicated questions and produce work-ready outputs.
The company says the new capability is aimed at users who need more than a quick answer. Deep Research in Computer can split a broad prompt into smaller tasks, route those tasks across more than 20 frontier models, and return formatted reports, presentations, and dashboards. Perplexity is positioning the feature as part of a broader workflow that keeps research, analysis, and document creation inside the same environment.
Perplexity says Deep Research in Computer is built on its Agent Search SDK and Search as Code architecture. In practice, that means the system can write code to assemble searches, carry out thousands of retrieval steps in parallel, and tailor the research process to the question at hand.
The company says the feature is designed to turn a complex query into a structured research plan. It then looks across hundreds of websites for primary sources and cites each claim it makes. Perplexity also says the system can incorporate a user’s own files and connected apps, allowing internal documents to be cross-referenced with live web data.
That combination is intended to support tasks such as market analysis, legal comparison, healthcare research, and financial evaluation. Perplexity highlighted example prompts ranging from analyzing AI chip company cash flow to comparing privacy laws across the United States and Europe.
The interface shown in Perplexity’s materials suggests a multi-step workflow. A user asks a question, the system gathers and compares data, a recommendation is drafted, and the finished output can then be shared with others. Perplexity says this sequence can continue within Computer, where the research can be turned into a deck or dashboard without exporting to another app.
Perplexity also emphasizes that Computer chooses from a large set of models depending on the task. The company says a legal reasoning model may be used for contract review, a data model for spreadsheet checks, and a writing model for final drafts. The goal is to match the model to the job instead of relying on a single system for every step.
In its promotional examples, Perplexity also points to a recommendation workflow for a coffee shop expansion analysis. The demo compares cities based on visitor traffic, rent, and competition, then generates a clear recommendation. Perplexity says the system previews changes before they are applied, so users can see edits before they stick.
Perplexity is also citing benchmark results to support the release. The company says Deep Research in Computer improves accuracy, depth of analysis, and citation quality. In its materials, it compares the system with a legacy version of Deep Research on several benchmarks, including Humanity's Last Exam, BrowseComp, and DeepSearchQA.
The examples shown indicate that Computer outperformed the older approach on those tests. Perplexity presented the results as evidence that the new version is better suited for difficult research tasks that require reasoning and source tracing.
The update reflects a broader industry push to combine search, reasoning, and productivity tools into agent systems that can do more of the preparatory work around an assignment. For Perplexity, Deep Research in Computer appears to be an effort to make that process more end-to-end, moving from question to cited output without leaving the platform.