HudHud Tech — Tech & AI news platform
Wednesday, 18 November 2026
Breaking
Artificial Intelligence

Perplexity Introduces Multiagent Feature for Enhanced AI Collaboration

18 November 2026 3 min read هدهد تك | Hudhud Tech

Perplexity has introduced Multiagent to its Computer platform, a new feature designed for research and multistep tasks. This feature represents a higher effort level than Ultra, enabling a lead agent to coordinate a team of up to 16 agents using different AI models. These agents work in parallel, share findings, and verify each other’s outputs, marking a significant structural change for developers and analysts.

TL;DR

  • Size: Up to 16 agents per task, plus the lead. Context length: Not disclosed.
  • Runs on: Perplexity’s cloud. Web only, Perplexity Max plan. No local or quantized option.
  • Performance: Not benchmarked. Perplexity has not published accuracy or speed metrics.
  • Best: 16-agent cap with direct agent-to-agent checks across three default model families (Claude, GPT, Gemini).

What is Perplexity Computer Multiagent?

Multiagent is the highest effort level available in Perplexity Computer, designed to coordinate AI models for analyzing complex tasks and cross-checking their outputs, as stated in Perplexity’s announcement.

This feature builds on the effort controls launched on September 17, 2026, which introduced four settings: Light, Standard, High, and Ultra. Each setting determines the orchestrator model and its reasoning depth, with lower settings utilizing less expensive models.

How does Multiagent work?

The lead agent is responsible for planning the task and assigning components to supporting agents, with the ability to assign up to 16 agents while also contributing to the work.

In an example provided by Perplexity, Claude Fable 5.2 serves as the lead agent, distributing tasks among agents running Claude Opus 5.5, GPT 6.1 Sol, and Gemini 4 Carbon, which form the default team. Claude Fable 5.2 and GPT 6 Astra act as supporting agents only when necessary.

Key mechanics from the launch post include:

  • Direct sharing: Agents utilize evidence discovered by others, challenge assumptions, and highlight discrepancies.
  • Shared files: Agents collaborate on edits to shared documents, allowing one agent to update revenue assumptions and notify others to incorporate them into a valuation.
  • Single point of contact: Users interact solely with the lead agent, who consolidates findings into a single response.
  • Follow-ups: The lead agent informs affected agents of any changes, allowing the team to continue their work seamlessly.
  • Control: Users can monitor which agents and models are active, and stopping the thread halts the entire team.
  • Model choice: Users have the option to specify particular models in their queries.

Multiagent vs Ultra vs Model Council

Perplexity now offers three methods for applying multiple models to a single problem:

  • Ultra: A single lead agent delegates tasks to supporting agents, which report only to the lead.
  • Multiagent: The same lead model as Ultra coordinates agents that communicate with each other and edit shared deliverables.
  • Model Council: Multiple models independently respond to the same question, with Computer displaying areas of agreement and disagreement. Model Council was introduced in July 2026.

Perplexity positions Multiagent as suitable for large, interconnected tasks that can operate in parallel, including deal data room diligence, multi-scenario forecasts, annual report reviews, competing-hypothesis research, and year-end tax preparation.

Related link: perplexity.ai

 

Source: MarkTechPost

What do you think of this story?

Share this story: