Oossa

Perplexity Computer adds Multiagent tier with up to 16 AI agents

Perplexity rolls out a new ‘Multiagent’ effort level that lets a lead agent coordinate up to 16 agents across Claude, GPT and Gemini models.

By Published by Oossa: 2 min read

Andrew Neel · Unsplash

Perplexity announced a new effort level for its Computer platform called Multiagent. The feature sits above the existing Ultra tier and lets a single lead AI run as many as 16 supporting agents at once. The agents operate on three frontier models – Claude Opus 5.5, GPT 6.1 Sol and Gemini 4 Carbon – and they share findings in real time, checking each other’s work before the lead assembles a final answer.

The lead model, Claude Fable 5.2, plans the overall task, splits it into subtasks and assigns each to a supporting agent. As the agents work, they can borrow evidence from one another, flag contradictory assumptions, or ask for more data. When all pieces are complete, the lead compiles a single response that reflects the combined analysis. Users stay in the conversation with the lead model; they do not need to message each agent separately. The new tier is offered to all Perplexity Max subscribers on the web and is selected via an effort slider in the Computer UI.

What kinds of work benefit from Multiagent?

Perplexity says the tier is aimed at tasks that can be broken into distinct parts and profit from parallel analysis. Example use cases include financial forecasting, where agents can model different revenue scenarios side by side, and data‑room diligence, where contracts, financial statements and operating documents are reviewed simultaneously. The platform also mentions tax‑return planning, annual report audits and deep research projects that need competing explanations examined in parallel.

How it differs from other effort levels

The Ultra level already delegates subtasks to supporting agents, but those agents only report back to the lead. Multiagent adds direct communication among agents, allowing them to cross‑check each other’s outputs before the lead sees them. A third option, Model Council, runs several models independently on the same question and then highlights agreement and disagreement, which is useful for pure judgment calls.

Why it matters

For a business analyst, Multiagent means a single query can produce parallel financial models and a consolidated report without juggling multiple tools. For everyday users, the new tier promises faster, more reliable answers on complex topics, though the quality still depends on the underlying models.

Was this article useful?
Share

Read next

Oossa · Newsletter

The week in AI, explained

Every Monday: the stories worth knowing, in plain language. Free, no spam.