# Perplexity Computer: The AI Agent Built to Run Entire Workflows Nikita Shrivastava Updated on: March 2nd, 2026 ⋅ Published on: February 26th, 2026 ⋅ 4 mins read > TL;DR Perplexity Computer is Perplexity AI's new multi-model agent platform that researches, codes, designs, and deploys projects end-to-end. Here's what it does, how it works, and why it matters. For two months, as OpenAI shipped new agents, Google folded Gemini into everything it touches, and Anthropic acquired a computer-vision startup to beef up Claude, Perplexity AI has been silent. Then, on February 25, 2026, it dropped Perplexity Computer, [a full-blown autonomous agent platform](https://www.perplexity.ai/hub/blog/introducing-perplexity-computer) that can research, design, code, deploy, and manage projects while you do something else entirely. The timing feels deliberate. As every major AI lab races to ship [agents that do things rather than just say things](/content/blogs/agentic-ai-vs-ai-agents/index.html), Perplexity has placed a specific bet: no single AI model should run your workflow alone. ## How Perplexity Computer Actually Works The workflow is simple. You describe an outcome like "build me a competitive intelligence dashboard," "create a monthly market report with charts," "design and deploy a landing page for my product launch.” The Perplexity Computer breaks that goal into subtasks. Each subtask gets routed to whichever AI model is best suited for the job. This is what Perplexity calls [multi-model orchestration](/content/blogs/ai-agent-orchestration/index.html), and this is what separates Perplexity Computer from most of what's on the market. The system currently runs 19 specialized models, including: - [Claude Opus 4.6 as the central reasoning engine](/content/blogs/claude-opus-4-6/index.html), deciding what needs doing and who does it - [**Google Gemini** for deep research](/content/blogs/gemini-3-1-pro/index.html) tasks - **Grok** for fast, lightweight queries - **Nano Banana** for image generation - **Veo 3.1** for video production - **ChatGPT 5.2** for long-context recall and broad web search Each task runs inside an isolated computing environment with access to a code interpreter, browser, file system, and hundreds of third-party connectors. The work is asynchronous. You can run dozens of parallel projects and check back when results surface. ## What Makes the Perplexity Computer Different from Other AI Agents? The AI agent market is getting crowded. [OpenClaw, the open-source rival](/content/blogs/open-claw/index.html) formerly known as Clawdbot, runs directly on your machine. Anthropic's Claude has its own agent framework. So what's Perplexity's angle? **Three things stand out:** - **Model flexibility over model loyalty.** Most agent platforms are locked to a single provider's models. Perplexity Computer treats models like interchangeable specialists, routing coding tasks to Claude, research to Gemini, and quick lookups to Grok. As newer models prove themselves, they get added to the rotation. - **Managed infrastructure.** Unlike open-source alternatives that run locally, Perplexity Computer operates entirely in the cloud on Perplexity's own infrastructure. That means centralized safety guardrails, automatic updates, and persistent memory across sessions. For enterprise buyers, that's a cleaner accountability story. - **Long-running autonomy.** This isn't a "prompt-and-respond" system. Perplexity says Computer can run workflows for hours, weeks, or even months, scraping data, generating recurring reports, and monitoring changes with minimal human check-ins. The tradeoff? You're trusting Perplexity's orchestration layer to make the right routing decisions every time. The company acknowledges that switching models mid-workflow can introduce inconsistencies and that long-running agents may drift as context windows fill up. ## Pricing: Credits, Caps, and the End of Flat-Rate AI Perplexity Computer is live now for Max subscribers ($200/month), with rollout to Pro and Enterprise tiers coming soon. Here's how the pricing breaks down: - 10,000 credits per month included with Max - 20,000 bonus credits at launch (valid for 30 days) - Usage-based billing, where you pick which models power your sub-agents and set spending caps - Users can control costs by routing cheaper tasks to lighter models This is the first time Perplexity has charged consumers per token. Your monthly AI bill could start looking less like a SaaS subscription and more like a cloud computing invoice, line items and all. The real question for teams isn't whether $200/month is expensive. It's whether the output justifies the token burn on projects that would otherwise eat hours of manual work. ## Early Limitations Worth Watching No product launch is without caveats, and Perplexity has been upfront about a few: - **No independent benchmarks yet.** There are no published audit results, third-party red-team reports, or standardized task success rates. - **Context drift on long workflows.** Agents running for extended periods may lose alignment with their original instructions. - **Orchestration complexity.** Routing across 19 models adds architectural overhead, and subtle inconsistencies can creep in when models hand off to each other mid-task. These aren't dealbreakers, but they're worth tracking as the product matures. ## The Bigger Picture Perplexity Computer is less a product announcement and more a thesis statement: the future of AI isn't about building the single best model. It's about building the best coordinator. Whether that thesis holds will depend on how well the orchestration layer handles edge cases, how transparently it manages costs, and how reliably it delivers on month-long tasks without drifting off course. But the ambition is unmistakable. Perplexity isn't trying to be your AI assistant anymore. It's trying to be your AI employee. And as these autonomous agents grow more capable, the question of how their outputs are presented to users becomes just as important as the outputs themselves. That's exactly the problem [Thesys](/content/site-root.html) is solving with [Generative UI, turning raw AI responses into interactive, context-aware interfaces](/content/blogs/generative-ui/index.html) that people actually want to use. ## FAQs 1. What is Perplexity Computer? Perplexity Computer is an autonomous AI agent platform launched on February 25, 2026. It breaks complex projects into subtasks and routes each to the best-suited model from a pool of 19 specialized AI systems, handling research, coding, design, and deployment from a single prompt. 2. How much does Perplexity Computer cost? Perplexity Computer is available to Max subscribers at $200/month, which includes 10,000 monthly credits. A 20,000-credit launch bonus is valid for 30 days. Pricing is usage-based with user-controlled spending caps. 3. Which AI models does Perplexity Computer use? Perplexity Computer orchestrates 19 models, including Claude Opus 4.6 (reasoning), Gemini (research), Grok (fast queries), ChatGPT 5.2 (long-context), Nano Banana (images), and Veo 3.1 (video), among others. 4. Is Perplexity Computer available to all users? Perplexity Computer is currently limited to Max subscribers on the web. Pro and Enterprise access is coming soon, though no specific date has been announced. 5. How is Perplexity Computer different from ChatGPT or Claude agents? Perplexity Computer uses multi-model orchestration. Rather than relying on one model, it routes each subtask to the most capable model available, treating AI systems as interchangeable specialists.