Perplexity AI has officially launched Perplexity Computer, a new cloud-based AI agent designed to automate complex workflows by orchestrating multiple advanced AI models. The innovative system, which became available on February 25, 2026, is accessible exclusively to Perplexity Max subscribers at a cost of $200 per month, plus additional credits for usage. The launch has generated considerable interest, with users like "luffy" expressing enthusiasm, stating in a recent tweet, "> excited to try Perplexity Computer! https://t.co/YMzwtSViqR".
Perplexity Computer operates as a general-purpose digital worker, integrating over 19 frontier AI models, including Claude Opus 4.6 for reasoning, Gemini for deep research, and GPT-5.2 for long-context tasks. It functions within an isolated Linux sandbox, offering a consistent cloud environment for executing tasks without local setup complexities. The platform is engineered for end-to-end execution, allowing users to describe an outcome and have the system break it down into subtasks, delegating to specialized sub-agents.
The system excels in generalist personal-assistant workflows, handling tasks suchs as research, report writing, scheduling, content creation, and data analysis. Its key differentiator is the "zero-setup" convenience, contrasting with highly customizable but setup-intensive alternatives like OpenClaw. Perplexity Computer also features persistent memory, allowing it to maintain context across sessions, and boasts over 400 managed OAuth connectors for various services.
Despite its advanced capabilities, early reviews highlight certain limitations, including potentially buggy connectors and a "black box" problem for developers, as the internal execution within the sandbox is not directly visible. The subscription model, coupled with additional credit consumption, also raises concerns about potential runaway costs for intensive use. Perplexity Computer aims to redefine how professionals interact with AI, moving from simple chat interfaces to comprehensive autonomous task execution.
