Perplexity Computer is a general-purpose digital worker that unifies Opus 4.6, Gemini, Grok, ChatGPT 5.2, Nano Banana, and Veo 3.1 into one autonomous system. You describe an outcome; it breaks it into tasks, spawns sub-agents, and executes everything asynchronously — with a real browser, a real filesystem, and real tool integrations. Available now for Perplexity Max subscribers.
What is Perplexity Computer?
Perplexity Computer is a general-purpose digital worker that operates the same interfaces you do. Think of it as an AI that doesn't just suggest — it does.
The distinction matters. Chat interfaces deliver answers. AI agents execute individual tasks. Perplexity Computer is a system that creates and executes entire workflows, capable of running for hours or even months without continuous human input.
"Frontier AI models are getting smarter. The best are becoming so capable that the products built around them are a bottleneck for showing their true potential." — Perplexity
The product is Perplexity's answer to that bottleneck: a harness powerful enough to let every top model do what it does best, coordinated in a single system under a single interface.
The three levels of AI — and where Computer sits
Computer doesn't replace chat or agents — it sits above them. When a workflow needs a quick web search, it spawns a Grok-powered sub-agent. When it needs to generate a 40-page research report, it delegates to Gemini's deep research capability. When it needs to produce a video summary, it calls Veo 3.1. The system decides. You describe the outcome.
How Perplexity Computer works, step by step
1. Describe an outcome
You don't write prompts. You describe what you want delivered: a market research report, a working web app, a competitive dataset, a pitch deck with sourced data. The level of specificity is up to you — Computer reasons about what's missing and creates sub-agents to fill gaps.
2. Task decomposition
Computer breaks your outcome into tasks and subtasks, mapping dependencies automatically. A document drafting task gets paired with a data-gathering sub-agent. A code deployment task gets paired with API key research and error-resolution sub-agents. The plan is created before execution begins — and you can review or adjust intermediate steps.
3. Parallel sub-agent execution
Each sub-agent runs in an isolated compute environment with access to a real filesystem, a real browser, and real tool integrations. One agent drafts while another researches. You can run dozens of Perplexity Computers in parallel across entirely separate workflows.
4. Asynchronous delivery
The work runs in the background. You focus on other things. Computer surfaces checkpoints only when it genuinely needs you — a missing credential, an ambiguous decision point. Otherwise, it works through to delivery. Workflows can run for hours or months.
When Computer runs into a problem, it creates sub-agents to solve it. It can find API keys, research supplemental information, code apps if necessary, and check in only if it truly needs you — not for every uncertainty.
The model orchestration layer: who does what
This is where Perplexity Computer makes its most technically significant bet. Rather than betting on a single model, it routes every subtask to the model best suited for that specific job.
| Model | Role in Computer | Why this model |
|---|---|---|
| Opus 4.6 | Core reasoning engine & orchestration | Highest reasoning capability for planning and delegation |
| Gemini | Deep research, spawning sub-agents | Best-in-class for long-horizon research tasks |
| Grok | Speed tasks, lightweight searches | Fast execution for high-frequency, low-complexity steps |
| ChatGPT 5.2 | Long-context recall, wide web search | Strongest at maintaining context across very long documents |
| Nano Banana | Image generation | Specialised image output within workflows |
| Veo 3.1 | Video generation | Frontier video synthesis for multimedia deliverables |
The model-agnostic harness means this list changes as models improve. You can also pin specific models to specific subtasks — giving teams control over cost, privacy, and capability trade-offs at a granular level.
Why Perplexity is uniquely positioned to do this
Perplexity's argument is that model specialisation — not commoditisation — is where AI is heading. As every frontier lab doubles down on its specific strengths, the most powerful system is the one that can intelligently orchestrate all of them.
Perplexity has been building toward this. Comet, the world's first AI-native browser, gave them a real browser layer. Comet Assistant established the personal agent foundation. Perplexity's deep research capability — which they claim is industry-best — gives Computer its research spine. Persistent memory and tasks infrastructure rounds out the stack.
Computer is where these pieces assemble into something new.
The historical parallel Perplexity is drawing
Perplexity reaches back to 1757 in their announcement. Mathematician Alexis Clairaut employed two human "computers" — a job title at the time, meaning an apprentice who performs calculations — to refine Edmond Halley's comet prediction. Working day and night for months, the trio split the work and predicted Halley's Comet perihelion within two days of accuracy.
The word "computer" has always meant the same thing: autonomous division of complex work, with accuracy as the central necessity. Perplexity is explicitly positioning its product as the modern realisation of that concept — AI as the computer, not just a tool on a computer.
What this means for marketing and business workflows
For teams currently managing multi-step workflows across disconnected tools, Perplexity Computer represents a structural shift in what a small team can execute. Consider what becomes possible:
- Competitive intelligence pipelines — describe a competitor monitoring workflow; Computer researches, compiles, formats, and delivers reports on a schedule
- Content at scale — from keyword brief to researched draft to formatted HTML, entirely automated, with citations
- Lead research and enrichment — feed it a list; it returns enriched profiles, LinkedIn summaries, and outreach angles
- Campaign reporting — connect APIs; get formatted performance reports drafted and delivered without a human touching a spreadsheet
- Multi-step publishing workflows — draft, fact-check, image, format, and stage content across platforms in one instruction
The unlock is the asynchronous, parallel nature. You're not waiting for one step to finish before the next begins. You're describing an outcome and getting back a finished deliverable.
Availability and pricing
Perplexity Computer is available now for Perplexity Max subscribers. Enterprise Max access is coming soon. The pricing model is usage-based, with Max subscribers receiving 10,000 credits per month. At launch, Perplexity is granting a one-time bonus of 20,000 additional credits — valid for 30 days after grant, available for existing users immediately and for new signups at account creation.
Users can set spending caps and choose which models power specific subtasks — giving cost control at the workflow level rather than just the account level.
The question the industry will be watching
The obvious question: if models commoditise, does the orchestration layer lose its edge? Perplexity's counter-argument is that models are specialising faster than they are commoditising — and that the ability to route intelligently across specialists will only become more valuable as the gap between general and specialised models widens.
Whether that holds depends on whether the big labs — Anthropic, OpenAI, Google, xAI — build equivalent harnesses themselves. They all have model depth. None yet has Perplexity's specific combination of search-native infrastructure, deep research capability, and model-agnostic orchestration.
That window may not stay open long. But right now, Perplexity Computer is the most complete implementation of the idea the industry has been circling for two years.