After the popularity of Claude Code and OpenClaw, AI search leader Perplexity officially launched its next-generation general AI agent system - Perplexity Computer this Wednesday. This product is defined as a "general digital worker" that can operate interfaces like humans and supports continuous task execution for hours or even months.
Differing from tools on the market that rely on a single model, Perplexity Computer adopts an innovative "model orchestration" strategy. It integrates 19 proprietary and open-source models from top laboratories around the world and automatically assigns them based on task specialties: for example, Opus4.6 handles logical sequencing, Gemini manages in-depth research, GPT-5.2 supports long-text memory, and Grok takes care of lightweight tasks requiring speed. This "each doing their own part" architecture makes automation of complex workflows possible.
AIbase learned that the system runs in a secure sandbox environment and has deep web access, code execution, and file generation capabilities. In internal testing, employees have used it to complete 4,000-line spreadsheets that would normally take a week of manual work in one night, and achieved full-process management from research, design to program deployment. Users no longer need to issue tedious step-by-step instructions; they only need to set the final goal, and the system will automatically break down the task and call multiple sub-agents to proceed in parallel.
Currently, Perplexity Computer is officially available on the web, initially opening to Max subscription users. Considering the high computing costs generated by automated agents, Perplexity has introduced a token-based billing mechanism for the first time. The launch of this product marks a significant leap in AI collaboration mode from "instruction-driven" to "goal-oriented," providing enterprises with a true digital labor solution.
Summary:
🤖 General Digital Laborer: Perplexity Computer is positioned as a general agent capable of creating and executing complete workflows, supporting integrated operations from research, programming to deployment.
🧩 Deep Multi-model Orchestration: The system integrates 19 top models including GPT-5.2, Gemini, and Opus, and uses intelligent task allocation to leverage each model's strongest strengths.
🎯 Goal-Oriented Collaboration: Users only need to describe the final result, and the system can automatically decompose the task and run it continuously for a long time, greatly improving efficiency in handling complex and heavy-load scenarios.
