Perplexity Computer: Multi-Model AI Workflow Platform Launches
Perplexity AI launched Perplexity Computer on February 25, 2026, as a unified multi-model AI workflow system designed to execute complex projects end-to-end. CEO Aravind Srinivas announced the platform after two months of development, describing it as a system that unifies files, tools, memory, and models into a single orchestrated environment. The platform operates as a general-purpose digital worker capable of running workflows for hours or months without constant supervision.
Unlike traditional chatbots that rely on single AI models, Perplexity Computer orchestrates 19 different AI models simultaneously. The system uses Anthropic’s Claude Opus 4.6 as its core reasoning engine while delegating specialized tasks to models optimized for specific functions. Users describe desired outcomes, and the platform automatically breaks tasks into subtasks handled by specialized sub-agents that coordinate asynchronously.
System Capabilities Overview
The system includes persistent memory, web browsing capabilities, file system access, and hundreds of connectors to external services. Each task runs in an isolated compute environment with access to a real filesystem, browser, and tool integrations. When problems arise, the platform creates sub-agents autonomously to find solutions, research supplemental information, or code applications as needed.
Multi-Model Architecture Breakdown
The model-agnostic architecture allows Perplexity Computer to swap models as new versions release and capabilities evolve. Users can manually override default model assignments for specific subtasks and set token spending caps to manage costs. This approach addresses the reality that no single AI model excels equally across all task types, from reasoning to visual processing to real-time web research.
How The System Operates
Access & Pricing Structure
20,000 one-time bonus credits for existing users and new signups (valid 30 days after issuance)
Platform Availability & Industry Context
Perplexity Computer became available to Max subscribers on February 25, 2026, through the web interface at perplexity.ai. The launch positions Perplexity in the autonomous AI agent market alongside competitors developing similar long-running task execution systems. CEO Aravind Srinivas framed the release as AI evolving from tool to platform, stating that when AI can orchestrate file systems, command-line tools, real-time internet access, and personal connectors, it essentially becomes the computer running operations in the cloud.
The system’s approach differs from locally-installed agent software by operating entirely within Perplexity’s cloud infrastructure rather than requiring direct access to user hardware. This architecture provides centralized management and safeguards while limiting potential security risks from over-permissioned local system access. The company described internal testing since January 2026, during which employees used Computer to publish documentation, build datasets, and create applications that previously required manual multi-day efforts.
The platform integrates with services including Gmail, Outlook, GitHub, Slack, Notion, and Salesforce through its connector infrastructure. As models continue advancing and specializing across different capabilities, the multi-model orchestration approach addresses the limitation that single AI systems cannot excel equally at all task types. Rollout to additional subscriber tiers will proceed based on platform performance under Max subscriber load.
