WorkGPT
Several people point pencils at printed charts on a meeting table
Photo: Photo by Pavel Danilyuk on Pexels

Perplexity Computer adds Multiagent, a team of up to 16 agents

Perplexity has added Multiagent to Computer, its agent for research and multistep work. Perplexity announced it on November 17. It is a new effort level above Ultra, and it is available to all Perplexity Max subscribers on the web.

With Ultra, one lead agent hands parts of a task to helper agents, and they report only to the lead. With Multiagent, the lead can assign up to 16 agents that work in parallel, share findings with each other, question assumptions and edit shared files. Claude Fable 5.2 plans the task. The default team mixes Claude Opus 5.5, GPT 6.1 Sol and Gemini 4 Carbon, and you can name other models in your request.

The lead stays your one contact in the chat. You can see each agent and its model while the task runs, and stopping the thread stops the whole team. For a follow-up, the lead tells only the affected agents what changed.

Perplexity's examples include financial forecasts with several scenarios, annual report reviews, data room checks and year-end tax planning.

What it means for your work

Multiagent suits a large job with parts that depend on each other, such as a forecast where revenue figures feed a valuation. For a question where you want separate opinions, Perplexity points to Model Council instead.

Give it a task you have already done by hand, so you can compare. Read the final report against your sources, because agents that read each other can also agree on the same mistake.

Perplexity also added phone calls to its teammates, covered in Perplexity teammates now take phone calls on iOS and Android.

Sources