Last updated: November 20, 2026
Perplexity Computer Multiagent, launched November 17, 2026, is a new effort level above Ultra. A lead agent assigns up to 16 agents running different frontier models, and they share findings and check each other’s work before the lead returns one answer. It is for Perplexity Max subscribers and uses more credits than Ultra.
Contents
- What changed on November 17
- How Multiagent works
- Availability, plans and price
- Ultra, Multiagent and Model Council compared
- How to run a task with Multiagent
- Workflows and prompt patterns
- Credits and cost control
- Limitations to plan for
- Frequently asked questions
What changed on November 17
Perplexity announced Multiagent in a blog post dated November 17, 2026. It adds a fifth step to the effort slider in Computer, above Ultra, for what Perplexity calls its most complex work.
Before, every effort setting had one lead agent delegating to supporting agents that reported only to the lead. In Multiagent, agents on different models work in parallel, share findings directly and check each other’s work, so conflicting assumptions surface before the lead merges the results.
How Multiagent works
The launch post describes the team:
- Lead and planning: Claude Fable 5.2 plans the task and divides it among the other agents.
- Team size: up to 16 agents, plus the lead doing part of the work.
- Default team: Claude Opus 5.5, GPT 6.1 Sol and Gemini 4 Carbon. Fable 5.2 and GPT 6 Astra join only when requested or needed.
- Model choice: you can name the models you want in your query.
- Direct sharing: agents reuse each other’s evidence, question assumptions and flag discrepancies.
- Shared files: agents coordinate edits, for example one updates revenue assumptions and tells another to use them in a valuation.
- One point of contact: you talk only to the lead, which merges the work.
While a task runs, the help center says an agent panel shows which agent is working on what. Stopping the thread stops the whole team. For follow-ups, the lead tells the affected agents what changed and the team continues its existing work.
Availability, plans and price
The effort mode help article says Multiagent is available to Perplexity Max subscribers in Computer on the web. On November 18, Perplexity posted on X that Teammates and Multiagent are now in the Perplexity app on iOS and Android for Max subscribers. Effort mode on mobile requires iOS app version 26.38.0 or later or Android version 2.100.0 or later.
Effort mode is not in the macOS and Windows desktop apps, which still show the regular model picker.
The Perplexity Max help page lists the plan at $200 per month or $2,000 per year. Our Perplexity Max cost answer has billing details.
On Enterprise, administrators set a maximum effort, default effort and maximum speed by role. Multiagent can be a role’s maximum but not its default, and it is hidden if the maximum is lower.
Ultra, Multiagent and Model Council compared
Perplexity now offers three ways to put several models on one problem. This table summarizes the official comparison and the effort mode documentation:
| Option | How the work is organized | Best for |
|---|---|---|
| Ultra | One lead delegates parts of a task to supporting agents, which report back only to the lead | Open-ended problems where several approaches need testing, such as a new market under different growth assumptions |
| Multiagent | Same lead model as Ultra; up to 16 agents share findings, check each other and edit shared deliverables | Large tasks with interconnected parts, such as data room diligence or a multi-scenario forecast |
| Model Council | Several models investigate the same question independently; Computer shows where they agree and disagree | Judgment calls that benefit from independent perspectives |
The practical difference from Ultra is cross-checking: a finding from one agent can change another agent’s analysis while the task runs.
For a comparable agent-team feature from Google, see our Antigravity /swarm guide. Perplexity’s default team mixes models from Anthropic, OpenAI and Google.
How to run a task with Multiagent
These steps follow the effort mode help article:
- Open Computer on the web, or in the Perplexity app for iOS or Android if Multiagent appears there for your account.
- Select the effort selector in the prompt box. On the web, it sits at the bottom of the prompt box, next to the current model name.
- Move the slider to Multiagent. If you do not see it, check that you are on Max, or ask your Enterprise administrator to raise your role’s maximum effort.
- Enter your prompt. Name specific models if you want them on the team.
- Submit, then watch the agent panel to see which agent is working on what.
- Send follow-ups to the lead. To stop all agents, stop the thread.
The setting is saved per browser or app and does not sync between devices. A change applies to your next submission, not to running work.
Workflows and prompt patterns
The announcement lists forecasting, annual report review, competing-hypothesis research, deal data room diligence and year-end tax planning as example uses.
These prompt patterns are our suggestions:
- Name the parts. “Build three revenue scenarios (base, downside, upside), a cost model and a valuation. Flag any assumption that differs between scenarios.”
- Ask for disagreements. “List every figure where the agents’ findings conflict, with the source behind each.”
- Name models only when it matters, for example GPT 6 Astra for one section.
- Define the deliverable, such as a spreadsheet or memo.
For ongoing work over days, see our Perplexity Teammates guide; for spoken prompts, see our Perplexity voice mode answer.
Credits and cost control
Computer runs on credits. The credits help article states that 100 credits equal $1 and gives typical ranges: 350 to 950 credits for a multi-step research report and 2,400 to 9,800 credits for a data dashboard. These are examples, not Multiagent prices.
Perplexity does not publish a per-task price for Multiagent. The effort mode article says it uses more credits than Ultra because a team of agents works on the task, and that no effort setting is a spending cap.
Our suggestion: test a prompt on High or Ultra first, then use Multiagent only where cross-checking matters.
Limitations to plan for
- Plan: Perplexity Max only.
- Desktop apps: not in the macOS and Windows apps.
- No cost ceiling: effort settings do not cap credits, steps, sources or tools.
- Running tasks: lowering the setting mid-task has no effect; restart the task.
- Pairings change: models behind each setting change over time.
- No accuracy figures in the documentation; judge results on your own tasks.
Frequently asked questions
What is Perplexity Computer Multiagent?
It is the highest effort setting in Perplexity Computer, launched November 17, 2026. A lead agent splits a task across agents on different frontier models that check each other’s work, then returns one answer.
Who can use Multiagent in Perplexity Computer?
Perplexity says Multiagent is available to Perplexity Max subscribers. It launched on the web, and Perplexity said on November 18 that it is now in the iOS and Android apps.
How many agents does Multiagent use?
The lead agent can assign up to 16 agents and also does part of the work itself. The default team combines Claude Opus 5.5, GPT 6.1 Sol and Gemini 4 Carbon, with Claude Fable 5.2 planning the task.
How many credits does Multiagent use?
Perplexity does not publish a fixed price per Multiagent task. It uses more credits than Ultra, and no effort setting is a spending cap. 100 credits equal $1.
How is Multiagent different from Model Council?
In Model Council, several models investigate the same question independently and Computer shows where they agree and disagree. In Multiagent, agents split one large task, share findings directly and contribute to shared deliverables.
Sources
- Perplexity blog: Introducing Computer’s Multiagent effort level (November 17, 2026)
- Perplexity Help Center: Effort mode in Perplexity Computer
- Perplexity Help Center: How credits work on Perplexity
- Perplexity Help Center: Perplexity Max
- Perplexity on X: Teammates and Multiagent in the iOS and Android app (November 18, 2026)
Image: Perplexity AI logo, public domain (PD-textlogo), via Wikimedia Commons, resized. Perplexity is a trademark of Perplexity AI, Inc. FreeAnswers is not affiliated with or endorsed by Perplexity.
Related guide: Replit Missions guide to agent swarms, budgets and the AI judge.
2 thoughts on “Perplexity Computer Multiagent: Power-User Guide to Agent Teams and Credits”