One dial in Perplexity Computer sets model and cost

Perplexity is rolling out an effort and speed selector for Computer, its agent for long multi-step work. Users now pick a preset instead of leaving every task on the same setup. The company’s own description, posted on Threads on September 17, says the presets “combine the orchestrator model and reasoning depth to control how deeply and efficiently Computer works through a task.”

The orchestrator sits at the top of the stack. Perplexity’s launch post describes how Computer splits an outcome into tasks and subtasks, then hands them to sub-agents for web research, document generation, data processing or API calls. When Computer launched on February 25, Opus 4.6 was the core reasoning engine. Other models covered narrower jobs: Gemini handled deep research, Grok took lightweight tasks where speed mattered, and ChatGPT 5.2 covered long-context recall. A preset now sets which model plans the work and how long it reasons before acting.

That choice also sets the bill. SQ Magazine reports that lower settings run on cheaper models, so what a user spends depends on the setting picked and the amount of work the agent does. It also says the custom controls are staying, so you can still name a specific model and set its reasoning level yourself.

For now the selector is on the web only. Android and iOS support is coming, with no date announced.