Perplexity has quietly turned a big corner: its Deep Research feature now lives inside Computer, the company’s ambitious “AI worker” that can plan and execute complex workflows over hours, even days. In practice, that means the same engine people have been using for long, citation-rich research reports is no longer just a button you press – it’s a capability that can sit inside an ongoing project, keep context, and keep working while you’re off doing something else.

If you’ve been following the generative AI race over the last couple of years, you’ve probably noticed how fast “chatbots” turned into “agents” and then into full-blown “AI employees.” Perplexity Computer is very much in that last category: not a single model, but a cloud service that orchestrates an entire roster of models – Claude Opus 4.6 for core reasoning, Gemini for heavy research, Grok for lightweight tasks, ChatGPT 5.2 for long-context recall, and specialized models for images and video. Deep Research was already one of Perplexity’s biggest differentiators, offering multi-step web research and structured reports on everything from finance to marketing; now, weaving it into Computer turns that feature from a one-off tool into a persistent research brain you can embed inside any workflow.

To understand why this matters, it helps to rewind a bit. When Perplexity first launched Deep Research, the pitch was almost disarmingly simple: instead of dumping a query into search, opening 15 tabs, skimming PDFs, and trying to synthesize everything yourself, you’d let an AI agent do it. Under the hood, Deep Research runs dozens of searches, reads hundreds of sources, iteratively plans what to look up next, and then distills everything into a report with inline citations and an exportable output you can turn into a PDF, document, or even a Perplexity Page. For non-subscribers, there’s a limited daily quota; for Pro and Max users, the feature effectively becomes a standard part of the toolkit, and Perplexity has been pretty vocal about treating this level of research capability as something everyone should have access to, not a luxury reserved for enterprise plans.

Deep Research also isn’t just a marketing label slapped on a slightly smarter search mode; Perplexity has tied it to measurable benchmarks. On SimpleQA, a benchmark for question-answering accuracy, Deep Research has hit around 93.9 percent, and on Humanity’s Last Exam – a notoriously difficult benchmark covering expert-level questions – it clocks in at about 21.1 percent, placing it among the stronger publicly discussed systems. Independent analyses comparing OpenAI, Perplexity, Google, and xAI’s deep research tools broadly agree on a trade-off: OpenAI’s Deep Research tends to lead on raw benchmark accuracy, while Perplexity is often faster, with many research runs finishing in under three minutes rather than five to thirty. That speed, plus the free tier and emphasis on transparent citations, has made Perplexity one of the more attractive options for people who care about both speed and verifiability, whether they’re students, analysts, or, frankly, journalists like you and me.

On its own, Deep Research already felt like a mini-agent: it planned, searched, reasoned, and wrote. But it still operated like a “mode” you turned on for a single question. Computer changes the frame. Perplexity describes Computer as a general-purpose digital worker – essentially a multi-agent system that can understand goals, break them into subtasks, pick the right model for each, and keep working with persistent memory, connected apps, and long-running workflows. Think of a system that can research a market, draft a product spec, spin up a prototype, monitor real-time data, and update your dashboards, all from a single starting prompt; that’s roughly the ambition, and Computer backs it up by orchestrating more than 19 models and running multiple search types – web, academic, people, image, video, shopping, social – in parallel.

Into that environment, Deep Research arrives not as a separate product but as a capability inside the stack. The official blog post announcing “Deep Research, now in Computer” is almost understated: it notes that Deep Research in Computer is available immediately, and that you can start a thread with a complex question and then continue working from the result, rather than treating every research run as a standalone artifact. In practice, that simple integration is powerful. It means you can kick off a Deep Research run on, say, “the competitive landscape of AI research tools for B2B SaaS,” then let Computer build on that report: spin up additional analyses, generate outreach lists, draft marketing collateral, or hook the findings into other workflows without ever losing context.

This tighter coupling also blurs a line that used to be fairly clear in the AI tools landscape. Traditionally, Deep Research-style features were bolted onto chatbots or search experiences: a “research” button in ChatGPT, a “long-form answer” in a search engine, a “report” toggle in some SaaS tool. Perplexity’s move with Computer effectively says: research isn’t just an answer type, it’s a primitive – something the system can call as a subroutine whenever it needs to answer complex questions, justify decisions, or keep itself updated with external data. That’s a shift towards what a lot of people in the AI space call “tools-using agents”: AIs that don’t just generate text, but actively decide when to search, when to read, when to code, and now, when to run a deep research cycle as part of a bigger plan.

For working professionals – especially in research-heavy roles – the impact is pretty immediate. A financial analyst can ask Computer to research a sector, synthesize trends, back up claims with data from sources like Statista or CB Insights, and then keep that context as the system builds models, dashboards, or even automated alerts, thanks to Computer’s new access to those data providers. A product marketer could use Deep Research to map out competitors, pricing strategies, and customer sentiment, then hand off to other Computer workflows that generate messaging frameworks, campaign calendars, or website copy, all while preserving the underlying research so you can trace a claim back to its sources.

The integration also lands in a broader competitive landscape where “deep research” has become a bit of a buzzword. OpenAI introduced its own Deep Research tool, positioning it as a higher-end capability targeted at professionals and pricing it accordingly, while Google has been experimenting with research-oriented modes in Gemini, and xAI’s Grok has explored search-heavy workflows of its own. Third-party comparisons note that Perplexity’s approach tends to emphasize transparency – inline citations, visible source lists, and a clear separation between what the system found and how it interprets it – whereas some competitors still feel more opaque, especially when they veer into speculative territory. The decision to make Deep Research available even on the free plan, albeit with limits, is a notable strategic contrast to OpenAI’s more locked-down model, creating a kind of “research for everyone” ethos versus a more premium-tier framing.

Of course, none of this magically solves the fundamental problems of AI-mediated research. Deep Research, like its peers, can still hallucinate, misjudge the credibility of a source, or gloss over uncertainty in ways that would make a human researcher wince. External evaluations stress that all current deep research tools require human oversight, especially when stakes are high: scientific work, legal analysis, healthcare, or major financial decisions. Perplexity’s own emphasis on citations and external benchmarks is partly an acknowledgment of that reality; by making it easier to audit where a claim came from and how the system got there, it invites users to treat Deep Research as a starting point, not a final authority.

What Computer adds to that equation is persistence and orchestration. Instead of treating every Deep Research session as a one-off project, Computer can keep running in the background, revisiting topics as new data comes in, updating reports, or kicking off follow-on tasks without waiting for a fresh prompt. It can integrate financial data, news, documents, and third-party tools into a single workflow, so Deep Research becomes one stage in a longer pipeline: research, plan, execute, monitor, and iterate. If Deep Research turned Perplexity into a kind of “supercharged search plus analyst,” then Deep Research inside Computer nudges it closer to a full research department – one that can think, read, write, and then actually do things with the knowledge it just assembled.

In the bigger arc of AI development, this feels like an early glimpse of what many companies are trying to build: systems that don’t just answer questions, but manage ongoing work. We’re still in the experimental phase; Computer is currently aimed at Perplexity Max/Pro subscribers and enterprise users, with a price tag and positioning that make it clear the initial target is power users and organizations rather than casual tinkerers. But features like Deep Research – especially when they’re exposed on both the consumer web interface and inside a more powerful “AI worker” layer – act as a bridge between today’s chat-centric AI experience and tomorrow’s long-running, goal-oriented agents.

If you’re the kind of person who already lives in research – whether you’re analyzing markets, writing long-form tech explainers, or sifting through documentation for a new product launch – Deep Research inside Computer offers a new workflow: you frame the problem, let the system do the initial heavy lifting, then stay in the loop as it turns raw information into plans, drafts, and actions. The catch is the same as with any powerful tool: you’ll get the most out of it if you treat it less like an oracle and more like a very fast, very capable junior colleague – one who can read everything, summarize anything, and keep working after you log off, but still needs your judgment to steer the story in the right direction.