Perplexity Launches Multi-Model AI System Perplexity Computer

Best known for AI search and Deep Research, Perplexity is positioning Computer as a digital worker that delivers finished results rather than merely answering questions. It can break down objectives and coordinate asynchronous subagents across research, design, coding, deployment and project management. The launch underscores how competition in AI agents is shifting from the capabilities of individual models to the orchestration of multiple models, tools and workflows.

Perplexity launched Computer on February 26, 2026, initially making the web version available to Max subscribers paying $200 a month. It can coordinate 19 models in parallel and includes 10,000 credits a month, plus a one-time bonus of 20,000 credits. On February 25, a user demonstrated an NVIDIA analytics terminal built in one afternoon, prompting discussion about whether it could replace the Bloomberg Terminal, which costs $30,000 a year. Its real-time data capabilities, however, remain less comprehensive.

Generative AI is evolving from one-off question answering into agent systems capable of carrying out tasks continuously. Pursuing a vision in which “everything is a computer,” Perplexity has launched Personal Computer, which connects disparate tools and handles cross-platform work on users’ behalf. New projects from Nvidia and Andrej Karpathy also reflect intensifying competition over AI agent infrastructure.

Issue 286 of AI Trends Weekly said Perplexity’s Personal Computer enables AI agents to work around the clock, 24 hours a day. Nvidia concurrently released Nemotron 3 Super, a model designed for agentic tasks, while OpenAI co-founder Andrej Karpathy open-sourced Autoresearch, an automated LLM research framework. No pricing or exact release dates were disclosed.

Perplexity’s Personal Computer brings multi-model agents to the Mac, connecting with local files, native apps and the Comet browser to carry out tasks across tools. Paired with an always-on Mac mini, it can operate 24/7 and is positioned as a digital twin for users who do not want to deploy OpenClaw themselves. Sandboxing, action audits, recovery tools and an emergency stop are designed to manage risk.

Perplexity unveiled Personal Computer at Ask 2026 on March 11, 2026, and rolled it out on April 16 to Max subscribers paying $200 a month. The app became available for Mac users to download on May 7, although the feature requires a Pro subscription costing $20 a month or a Max plan. It supports macOS 14 or later and can orchestrate more than 20 models. A Windows version has not been released.

Perplexity Computer is a cloud-based, multi-model task agent developed by Perplexity. Instead of relying on a single model to answer questions, it breaks research, design, software development and deployment into workflows handled collaboratively by different models and subagents. The product marks a further shift in generative AI from chat tools toward systems capable of operating autonomously for extended periods.

Perplexity unveiled Computer on February 26, 2026. It can orchestrate 19 AI models in parallel and connect to hundreds of services, enabling subagents to browse the web, write code and operate applications. The product was initially made available to Max subscribers paying $200 a month, with 10,000 monthly credits and a one-time bonus of 20,000 credits.

Perplexity AI began as a generative search provider and is now expanding into autonomous agents and workflow automation. Computer is not a single model. It selects and coordinates different AI models for tasks such as writing, coding and image generation, with the goal of reducing manual work and competing with the open-source autonomous agent OpenClaw.

Perplexity AI unveiled Computer in July 2026, positioning it as an orchestration system for frontier AI models. The product can break down complex tasks, select suitable models and continue executing the work. The company has not disclosed pricing, subscription fees or a date for general availability. Attention is now focused on its automation performance, barriers to adoption and future commercialization plans.