Story: Perplexity has introduced Perplexity Computer, a new multi-model orchestration system that dispatches work across 19 separate AI models, framing model choice and routing as a first-class product capability rather than a hidden backend detail. (Product page)

How It Works: Outcome → Sub-Agents → Execution

Users describe an outcome, and the system spins up sub-agents that can browse, code, connect to apps, and autonomously complete tasks end-to-end. The core promise is not just “chat,” but durable execution — turning requests into multi-step workstreams.

Multi-Model Routing as a Product Feature

Each job runs in its own sandbox and can freely mix and orchestrate rival models across tasks. Perplexity is explicitly leaning into a “best model for the job” worldview — with users able to hand-pick which model tackles which step when needed.

Always-On Jobs and Long-Running Work

Perplexity claims jobs can run actively for long durations (even months), which signals a shift from short, session-based assistants toward persistent agents that keep context and keep going.

The Competitive Shot: Walled Gardens vs Model Flexibility

CEO Aravind Srinivas also took a direct shot at single-vendor ecosystems, arguing that a major weakness of closed assistants is that they only “cowork” with their own models. Perplexity Computer is the opposite bet: orchestration across many.

Pricing and Controls

Pricing is described as consumption-based, with higher tiers receiving a monthly credit bank (e.g., a Max tier with a larger bank). The positioning here is clear: as agents do more work, pricing follows usage — not just a flat “chat subscription.”

Why It Matters

Multi-model choice has been creeping into AI products, but Perplexity Computer is pushing it into the center: a durable, sandboxed, agent-style system where flexibility is the interface. If this works reliably, it strengthens the argument that the “winning assistant” may be the one that orchestrates best — not the one that owns a single best model.