← Tuesday, November 17 deck

35Product2 sources posting

Perplexity Rolls Out Multiagent, Where a Lead Agent Splits Work Across Agents Running Different Frontier Models

We’re rolling out Multiagent, a new effort level in Computer and teammates. A lead agent splits tasks across a team of agents using different frontier models. As they work, agents share findings with one another, with the lead agent combining everything into one answer.

Opening frame from Perplexity's 41 second video: dark serif text reading Introducing Multiagent on a plain off white background, with the second word partly animated in.

Context

Why it matters

Letting one lead agent split a task among agents built on models from different labs turns model choice into a team setup, which Perplexity says makes complex work faster and more accurate.

Perplexity posted that Multiagent is a new effort level in its Computer and teammates products. CEO Aravind Srinivas added that it is available to all Max subscribers on the web, with mobile support coming soon. The same day Perplexity also launched its Orchestrator API, covered separately in this deck.

Deep dive

Perplexity announced that it is rolling out Multiagent, a new effort level in Computer and teammates. According to the post, a lead agent splits a task across a team of agents that use different frontier models. As they work, the agents share findings with one another, and the lead agent combines everything into one answer.

Aravind Srinivas, Perplexity's CEO, gave more detail in his own post. He said users can turn it on through the Effort level setting and gave an example in which Fable 5.2 orchestrates Gemini 4 Carbon, GPT 6.1 Sol and Opus 5.5 in parallel agent streams, to get complex tasks done faster and more accurately. He named financial research, forecasting, prediction markets and analyzing a batch of filings together as strong uses. He said it is available to all Max users on the web, with mobile support coming soon.

The speed and accuracy claims are Perplexity's own and come without published benchmarks in the posts reviewed here. The posts also do not say how many agents run at once or how usage is counted against plan limits.

The launch came on a busy day for the company. Perplexity also introduced its Orchestrator API, which picks models and providers for developers, and Srinivas said the company had passed $1 billion in annual recurring revenue. Both are covered elsewhere in this deck.

What to expect next

Watch for mobile support, which Srinivas said is coming soon, and for whether Multiagent reaches plans below Max. Independent comparisons against single model agents on research tasks would test the accuracy claim. Rival assistants that already route between models may answer with similar team features, and the cost of running several frontier models per task will shape how widely it is offered.