Key Takeaways
- Perplexity Computer launched February 25, 2026, exclusively for Max subscribers ($200/month or $2,000/year).
- It coordinates 19 AI models simultaneously, routing each subtask to the best-suited model.
- Claude Opus 4.6 serves as the core reasoning engine; Gemini handles deep research; Nano Banana generates images; Veo 3.1 produces video; Grok covers fast, lightweight tasks; GPT-5.2 manages long-context recall.
- Every task runs in an isolated cloud environment with a real filesystem, browser, and 400+ app integrations (Slack, Gmail, GitHub, Notion).
- Max subscribers receive 10,000 credits per month; tasks consume credits based on complexity.
- Enterprise Max is available at $325/seat/month with added security controls, audit logs, and SSO.
- A separate “Personal Computer” product runs locally on a Mac Mini for users who want on-device AI with local file access.
Perplexity Computer is an autonomous AI agent launched on February 25, 2026, that breaks complex tasks into smaller pieces and assigns each one to whichever AI model handles it best.
Available through the $200/month Perplexity Max subscription, it orchestrates 19 specialized AI models — including Claude Opus 4.6, GPT-5.2, Gemini, and Grok — inside a secure cloud sandbox, running workflows that can last hours, days, or even months without constant human oversight.
Think of it as a project manager that never sleeps. A user describes a desired outcome — “build me an interactive stock dashboard” or “plan and execute a digital marketing campaign for my restaurant” — and Computer decomposes that goal into subtasks, spawns specialized sub-agents, assigns each one to the right model, and delivers finished results. It can search the web, generate code, produce images and video, draft documents, and interact with over 400 connected apps, all in parallel.
Why Perplexity Built a Multi-Model AI Agent
Most AI products today rely on a single model. Perplexity’s argument is that frontier AI models are not becoming interchangeable generalists — they are specializing. Each model excels at different types of work: some are better at reasoning, others at coding, visual tasks, or speed. A full workflow that spans research, writing, design, and deployment needs access to multiple models and the intelligence to pick the right one at each step.
Perplexity’s own enterprise data backs this up. In January 2025, 90% of queries across their enterprise customer base routed to just two models. By December 2025, no single model commanded more than 25% of usage. That usage pattern told the company where the market was heading: toward orchestration, not loyalty to a single provider.
As CEO Aravind Srinivas put it: “I don’t think Computer could have been as powerful as it is if we had launched it even three months ago. You have to have the right harness at the right time.”
How Perplexity Computer Works Step by Step
The system follows a clear sequence. A user provides a high-level objective in natural language. Perplexity Computer’s central reasoning engine — currently running Claude Opus 4.6 — parses the request and maps out a plan of tasks and subtasks. It then spawns sub-agents, each assigned to the model best suited for that particular job.
These sub-agents work asynchronously. While one agent drafts a document, another gathers the data that document needs. A third might generate images. A fourth could be running code in a sandboxed environment. The user does not need to manage any of this. Computer checks in only when it genuinely requires human input.
If a sub-agent hits a problem — a missing API key, insufficient context, a code error — Computer creates additional sub-agents to troubleshoot. It can search for supplementary information, write custom code on the fly, or pivot its approach. The entire process runs inside an isolated cloud environment with 2 vCPUs, 8 GB of RAM, Python and Node.js pre-installed, and access to a real browser and real file system.
Which AI Model Does What
| Model | Role Inside Computer |
|---|---|
| Claude Opus 4.6 (Anthropic) | Core reasoning engine, orchestration logic, coding |
| Gemini (Google) | Deep research, creating sub-agents for complex queries |
| Nano Banana | Image generation |
| Veo 3.1 (Google) | Video production |
| Grok (xAI) | Fast, lightweight tasks where speed matters |
| GPT-5.2 (OpenAI) | Long-context recall, wide web search |
Perplexity’s platform is model-agnostic by design. As newer or better models appear, they can be swapped into the orchestration pipeline. Users can also manually choose specific models for specific subtasks if they want tighter control over token budgets and output quality.
What Can Perplexity Computer Actually Do?
Perplexity positions Computer as a general-purpose digital worker. In practice, early adopters and the company have demonstrated several categories of work.
On the research side, Computer can generate detailed competitive intelligence reports, build investor lists, compile market analysis, and create structured spreadsheets — all from a single prompt. It runs up to seven search types simultaneously (web, academic, images, and more), reading full source pages rather than brief snippets.
For coding and design, it can build, test, and deploy functional applications. Early users have had it build complete data visualization websites, interactive stock trackers, and financial dashboards resembling Bloomberg Terminal layouts. It writes code in its sandboxed environment, tests it, debugs errors, and iterates until the output works.
On the operational side, Computer handles project management tasks, schedules appointments, sends emails through connected services, updates databases, and monitors competitors on a recurring basis. With over 400 managed OAuth connectors — Slack, Gmail, GitHub, Notion, and many more — it can interact with the same tools teams already use.
Pricing, Credits, and What It Actually Costs
Perplexity Computer is not a standalone product. It ships as part of the Perplexity Max subscription at $200 per month ($2,000/year if billed annually). The Max plan also includes unlimited Pro searches, access to all advanced models, Sora 2 Pro video generation, the Comet AI-native browser, and unlimited Labs usage.
Computer runs on a credit-based system. Max subscribers receive 10,000 credits per month. Each task drains credits based on its complexity — a simple alt-text generation might cost around 30 credits, while a large coding session can burn thousands. At launch, new and existing users received a one-time bonus of 20,000 to 35,000 extra credits (reports vary), valid for 30 days.
| Plan | Monthly Price | Monthly Credits | Computer Access |
|---|---|---|---|
| Perplexity Pro | $20 | Coming soon | Not yet |
| Perplexity Max | $200 | 10,000 | Yes |
| Enterprise Max | $325/seat | 15,000/seat | Yes |
When credits run out, active tasks pause rather than cancel. Progress is saved, and work resumes once credits become available — either through a monthly reset or a manual top-up. An auto-refill feature exists but is turned off by default, with a spending cap (default $200, adjustable up to $2,000) to prevent surprise bills.
One notable gap: Perplexity has not published a per-task credit breakdown. Users can see what each task cost after it runs, but there is no public table showing expected costs for different workflow types. Early testers have reported that complex multi-step automations and large codebase scans can consume credits faster than expected — one Reddit user reported a 280,000-line Python codebase scan burning through 21,000 credits.
Perplexity Computer vs. OpenClaw vs. Claude Cowork
Perplexity Computer enters a field that already includes OpenClaw — a viral open-source agentic tool — and Anthropic’s Claude Cowork. Each takes a different approach to the same underlying idea: letting AI agents do sustained, autonomous work.
OpenClaw runs locally on a user’s machine. It offers extreme flexibility and has driven rapid adoption among developers, but it also puts the burden of configuration and security on the user. It relies on an unregulated ecosystem of plugins, which has led to documented incidents, including one case where the tool deleted a user’s emails without authorization. OpenAI later hired OpenClaw’s developer, with CEO Sam Altman saying some of what OpenClaw demonstrated would be essential to OpenAI’s product vision.
Claude Cowork, from Anthropic, focuses on enterprise collaboration but only uses Anthropic’s own models. That means it lacks the model diversity Perplexity offers.
Perplexity Computer sits between the two. It runs in the cloud, not on a local machine, which simplifies setup and reduces the security risks of local file access. It uses a curated integration ecosystem — more like an App Store than the open web, to borrow the analogy offered by Ars Technica. And its multi-model orchestration means it is not locked into any single provider’s strengths or weaknesses.
| Feature | Perplexity Computer | OpenClaw | Claude Cowork |
|---|---|---|---|
| Execution | Cloud-based | Local machine | Cloud-based |
| Models used | 19 models, multi-provider | User-configured | Anthropic models only |
| Integrations | 400+ managed OAuth | Unverified plugins | Curated, MCP-based |
| Setup required | None (managed) | High (self-configured) | Low |
| Pricing | $200/month (Max plan) | Free / open source | Included in Enterprise plans |
Personal Computer: The Local Hardware Option
At its inaugural Ask 2026 developer conference on March 11, Perplexity unveiled a companion product called “Personal Computer.” This is separate software that runs on a dedicated local device — such as a Mac Mini — giving the cloud-based AI agent persistent access to local files, apps, and sessions.
Perplexity says Personal Computer is more secure than OpenClaw because it requires all actions to be confirmed by the user and includes a built-in audit trail. The two products are designed to complement each other: Computer runs complex workflows in the cloud, while Personal Computer bridges the gap to a user’s local environment.
Enterprise Expansion and Market Context
Just two weeks after the consumer launch, Perplexity made Computer available to enterprise customers at the Ask 2026 conference. Enterprise Max costs $325 per seat per month ($3,250/year) and adds organization-level security controls, compliance features, audit logs, SCIM provisioning, configurable data retention, and Slack integration — employees can query @computer directly inside Slack channels.
The enterprise push comes at a time when demand for agentic AI is surging. A February 2026 survey by CrewAI found that 100% of surveyed enterprises plan to expand their use of AI agents this year, and Fortune Business Insights projects the global agentic AI market will grow from $9.14 billion in 2026 to $139 billion by 2034. Perplexity says more than 100 enterprise customers contacted the company over a single weekend demanding access after seeing early user demonstrations on social media.
Perplexity itself is valued at roughly $20 billion following its Series E-6 funding round. Its annual recurring revenue grew from $80 million in late 2024 to an estimated $200 million by February 2026, with internal projections targeting $656 million by the end of 2026.
Risks and Limitations Worth Knowing
Perplexity Computer is not without trade-offs. The credit system lacks transparency — there is no published table mapping task types to credit costs, making it difficult to budget accurately. Heavy users have reported spending $300 to $500 per month or more beyond the base subscription.
LLMs still make mistakes. If Computer is working with data a user hasn’t backed up, or if outputs go unverified, errors could have real consequences. The cloud-based approach also means sensitive data passes through Perplexity’s infrastructure, a concern that enterprise security teams will need to evaluate carefully.
The product is also not ideal for software developers who need visual feedback loops, live previews, and deep environment control. For casual users, the $20/month Pro plan covers most standard research and search needs without the overhead of a credit system.
Perplexity also faces ongoing legal challenges. Several copyright lawsuits are pending against the company, and investigations by Wired and Cloudflare have documented its use of undisclosed web crawlers that bypass websites’ scraping restrictions.
The Bottom Line
Perplexity Computer is an ambitious bet that the future of AI productivity lies not in any single model, but in the intelligent orchestration of many. For professionals whose work spans research, content production, coding, and multi-tool coordination, it consolidates capabilities that previously required juggling multiple AI subscriptions and manual workflows.
The $200 entry price and opaque credit system will give some users pause, but for those who need a managed, multi-model AI agent that can work autonomously for hours on end, Computer is currently the most fully integrated option on the market.
If you are interested in this topic, we suggest you check our articles:
- How Much Does Perplexity Computer Cost? Full Pricing, Credits, and Plan Breakdown
- How Does Perplexity Compare Against Other GenAI Models?
- 6 Tasks Better Suited to Perplexity Than Other AI Tools
Sources: Perplexity, Ars Technica
Written by Alius Noreika
