Perplexity launches Computer routing tasks across 19+ AI models

Perplexity unveiled Computer this week, a cloud-based orchestration system that accepts high-level objectives, decomposes them into subtasks and routes each piece to the model best suited to the job. The company says the platform “unifies every current AI capability into a single system” and bills Computer as “a general-purpose digital worker” that “reasons, delegates, searches, builds, remembers, codes, and delivers.”

Perplexity made Computer available immediately to Perplexity Max subscribers and said it will roll the product out to Pro and Enterprise customers “in the coming weeks.” The company also announced a developer event on March 11 to disclose more details about its search API, ranking embeddings and the infrastructure behind its orchestration stack.

Under the hood, Computer is a multiagent, model-agnostic layer that spawns sub-agents to tackle components of a task. Reporting identifies a roster of frontier models wired into the system, explicitly naming Claude, Gemini and Grok among others, and cites specialized models for images and video. Those reports list Anthropic’s Opus 4.6 as a core reasoning engine, Google’s Gemini for deep research, ChatGPT 5.2 for long-context recall and broad searches, Grok for fast lightweight tasks, Nano Banana for images and Veo 3.1 for video. Sources vary in the exact tally; Perplexity’s materials and outside coverage describe the platform as drawing on more than a dozen, and in some counts 19 or more, frontier models.

Perplexity frames the product as a strategic bet that the value in AI will accrue to the orchestration and abstraction layer rather than to any single monolithic model. The company cites enterprise usage data showing a shift “from two-model dominance to no-model dominance in just 12 months” to justify building a neutral layer that can swap in new models as they improve.

To address security and operational concerns, Perplexity runs Computer’s tasks in isolated compute environments inside its secure cloud sandbox, with access to a filesystem, a browser and tool integrations. The system is designed for long-running workflows, Perplexity claims Computer can run for hours or even months, and users can run multiple instances in parallel. Perplexity says the sandbox approach avoids the credential and local-setup complexities that have plagued open-source, always-on agents.

The launch comes amid a broader debate about autonomous agents. Developers of viral local agents have demonstrated powerful automations but also exposed new attack surfaces, and major platform players have moved quickly to hire talent from that ecosystem. OpenAI’s chief executive publicly praised an open-source agent developer as bringing “amazing ideas about the future of very smart agents interacting with each other to do very useful things for people,” signaling the industrywide interest in agent models.

Safety experts caution that orchestration does not eliminate risk. Perplexity’s cloud boxing reduces some vectors associated with local agents, but large language models still make mistakes and errors could be consequential if long-running workflows act on unbacked data or carry out unchecked actions. The company will face questions about audit logs, credential handling, data retention and third-party security validation as enterprises begin to adopt Computer for production tasks.

Perplexity positions Computer as an abstraction layer much like early cloud middleware, arguing that companies that build the best orchestration above commodity models will capture disproportionate value. With a March 11 developer event on the horizon and a staged rollout to paid tiers, Perplexity is making a practical case for multi-model workflows while pushing the industry conversation from model supremacy to orchestration and governance.