Perplexity Computer Adds New Multiagent Feature: Up to 16 Agents Cross-Check Each Other's Work
Perplexity Computer has launched a new paid tier called Multiagent, positioned above the existing Ultra tier. One master agent can orchestrate up to 16 agents running on Claude, GPT and Gemini, which share their findings and perform cross-verification. Complex tasks gain an extra layer of parallel processing and validation, at the cost of longer runtime and higher API call expenses.
Perplexity has rolled out a new tier for its Computer product: Multiagent. Placed above the Ultra tier, its core change allows a master agent to schedule up to 16 agents that run on Claude, GPT and Gemini models respectively, share their discoveries, and check each other's work.
This essentially turns a single question-and-answer session into a small collaborative workflow. The master agent is in charge of breaking down tasks, assigning jobs, and collecting final results; sub-agents run in parallel to look up information, test solutions, and perform validation. For tasks that are prone to missing details, such as research, code troubleshooting, and long document verification, the value of the multiagent setup lies in parallel exploration and cross-validation. The tradeoffs are direct: more model calls, longer waiting time, and more complex scheduling and result merging. The original source does not elaborate on pricing, latency and regional availability.
What is worth noticing is that Perplexity did not package Multiagent as a single model upgrade, instead it puts Claude, GPT and Gemini under the same scheduling layer. Simply increasing the number of models does not solve the problem, the key lies in scheduling logic and shared context.
Source: MarkTechPost, by Michal Sutter.
发布时间: 2026-11-18 06:52