Perplexity announced Multiagent for Computer on November 17, 2026. The new effort level lets a lead agent divide connected work among as many as 16 agents, which can exchange findings and check one another’s work before the lead assembles a response, according to Perplexity. The lead can also handle part of the task itself.

How Multiagent coordinates a task

Multiagent is designed for assignments with distinct but related parts. Rather than having several models independently answer the same question, a lead agent delegates work that can proceed in parallel, while supporting agents share findings and flag discrepancies for further investigation. They can also coordinate edits to shared files and pass updated inputs between analyses.

Multiagent fits work whose separate parts can inform one another; Model Council is for independent perspectives on the same question.

Perplexity says users choose the mode in Computer’s effort slider, can request models in a prompt, and can see which agents and models are active. The lead remains the user’s point of contact, and stopping the thread stops the team. NeoTeo previously covered Perplexity Computer’s fully local mode on Mac, a separate development focused on where processing runs.

Which models lead and support the work?

Perplexity names Claude Fable 5.2 as the planner and lead. The default supporting models are Claude Opus 5.5, GPT 6.1 Sol, and Gemini 4 Carbon. Claude Fable 5.2 and GPT 6 Astra can also serve as supporting agents when requested or when a task component calls for them.

How Multiagent differs from Ultra and Model Council

The three modes organize model work differently. In Ultra, supporting agents report back to the lead. Multiagent lets agents exchange findings across connected parts of an assignment. Model Council has several models investigate the same question independently before Computer combines their perspectives.

That makes the choice a matter of task structure: connected work that benefits from agents sharing information suits Multiagent’s stated design, while separate answers to one question fit Model Council’s approach.

What work Perplexity says Multiagent is designed for

Perplexity lists financial forecasting, reviewing annual reports, investigating competing explanations, deal data-room diligence, and complex tax-return and year-end planning as intended uses. For tax-related work, the company describes preparing analysis and questions for a tax professional.

Perplexity says Multiagent is available on the web to all Perplexity Max subscribers.