Perplexity adds 16-agent Multiagent mode to Computer
Perplexity’s new Multiagent effort level for Computer lets up to 16 agents share findings, edit common files and work through one lead. It is available on the web to Perplexity Max subscribers.
Perplexity has introduced Multiagent, a Computer effort level that coordinates up to 16 agents on one assignment. The company positions it above Ultra and says supporting agents can exchange findings directly, check one another’s work and edit shared files, while a lead handles the user conversation.
With Multiagent selected, the lead can assign work to up to 16 agents and take on part of the task itself, according to Perplexity. The company says Claude Fable 5.2 plans and divides the assignment by default. Claude Opus 5.5, GPT 6.1 Sol and Gemini 4 Carbon are among the supporting models, and users can request other supported models in their prompt.
That direct communication distinguishes Multiagent from Computer’s Ultra effort setting. Ultra relies on one orchestrating lead to choose an approach, delegate portions of the job and collect results from supporting agents. Multiagent uses the same lead model, Perplexity says, but allows supporting agents to share evidence, challenge assumptions, flag discrepancies and coordinate changes to a common deliverable. The new option joins Perplexity’s existing Computer controls as a separate effort choice.
Multiagent also takes a different approach from Model Council in Computer. Model Council sends the same question to two to eight user-selected models independently, then synthesizes areas of agreement, disagreement and unique findings. Multiagent instead divides an interconnected assignment into parts that can move forward in parallel, with agents communicating when their work overlaps.
Users steer the team through the lead rather than messaging each agent separately. Perplexity says the lead combines the team’s input into the requested response and files, then sends follow-up changes to the affected agents while the rest continue their work. Users can see which agents and models are active. Stopping the thread stops the whole team.
Multiagent is available on the web to all Perplexity Max subscribers. It can be selected from Computer’s effort slider when starting a new task.
More news

Perplexity launches Frontier Research, claims access to 10,000-plus GPUs

Researchers train fixed AI-agent teams to reason together

Perplexity explains how SPACE routes agent requests to sandbox executions
