Perplexity Computer: The AI Agent That Coordinates 19 Models to Get Work Done

Published: March 2026

March 202612 min read

From Search Engine to Digital Worker

Perplexity AI built its name as an intelligent search engine: ask a question and get a cited answer. That reputation helped it reach a $20 billion valuation. But on February 25, 2026, the company launched what CEO Aravind Srinivas called the most ambitious product in its three-year history: Perplexity Computer. [1]

This is not just a chatbot upgrade. Perplexity describes Computer as a system that accepts high-level objectives, decomposes them into subtasks, and delegates those subtasks to whichever model among its 19 specialists is best suited for the job. The strategic thesis is clear: win the orchestration layer above model vendors. [3]

What Perplexity Computer Actually Does

Perplexity positions Computer as a general-purpose digital worker. Users provide a goal, and the system plans, executes, and delivers outcomes through software interfaces in cloud sandboxes, often with minimal human intervention. [2]

  • Analyze large spreadsheets and generate summary reports
  • Build and deploy websites from a single prompt
  • Run competitive research and produce presentations
  • Plan multi-day travel with booking support

The distinguishing trait is the pairing of long-running execution with cross-platform chaining. Workflows can run in the background for hours or longer and can be parallelized across multiple Computer instances. [4]

The 19-Model Architecture

Perplexity challenges the idea that one giant model should do everything. Its operating data suggested specialization performs better in production. In Srinivas's framing, specialized models become tools in a broader system.

ModelPrimary Role
Claude Opus 4.6Core orchestration logic and coding
GeminiDeep research and sub-agent creation
GPT-5.2Long-context recall and broad web search
GrokSpeed-sensitive lightweight tasks
Veo 3.1Video generation
Nano BananaImage generation
+13 additional modelsDomain-specific subtasks

Perplexity says this roster will remain fluid and users can override the orchestrator to pin specific models for specific subtasks. [1] [3]

Three Capabilities Beyond Chat

  • Real interface operation: it clicks through web UIs, fills forms, downloads files, and executes workflows.
  • Cross-platform chaining: one prompt can gather web data, analyze it in sheets, and produce final slides.
  • Long-running async execution: goal-oriented runs continue in the background and request human input only at decision points.

Safety Architecture: Sandboxing by Design

The launch came amid increased concern over autonomous agent safety, especially after the OpenClaw incident where an agent deleted hundreds of emails after safety instructions degraded during context compaction. Perplexity's response is strict isolation: each task runs in a dedicated sandbox with separate filesystem and browser session so user data stays isolated from agent execution. [2]

Pricing, Availability, and Strategic Implications

Perplexity Computer is currently available to Perplexity Max users at $200/month or $2,000/year, with planned expansion to Pro subscribers in later phases. Perplexity frames this as paying for a tireless digital worker that handles multi-step projects with minimal oversight. [4]

The bigger thesis is durable: the strongest AI moat may belong to the company that orchestrates best-in-class models into dependable outcomes, not necessarily the company that builds any one model.

Sources