Most AI assistants hand your question to one model and return one answer. Perplexity's new Multiagent takes a different route: it puts a whole team of AI agents on the task, lets them check each other's work, and then gives you a single answer. Perplexity launched Multiagent on 17 November 2026, above Ultra in Perplexity Computer, its agent platform for research and multistep work.
Here are five things to know.
1. A team of up to 16 agents works on one task
Multiagent can divide a task among up to 16 supporting agents, each running on a different frontier model, working in parallel and sharing findings directly with each other as they go, rather than working in isolation and reporting back at the end.
That parallel structure matters most when a task has several distinct parts. Instead of one model working through a long job step by step, the parts can be handled at the same time.
2. A lead agent plans the work — and does some of it too
Every Multiagent run starts with a lead agent. The lead plans the task, breaks it into pieces, and hands those pieces to the supporting agents. In the default team, the lead runs Claude Fable 5.2, while the supporting agents run Claude Opus 5.5, GPT 6.1 Sol and Gemini 4 Carbon.
Claude Fable 5.2 and GPT 6 Astra can also join as supporting agents when the user requests them, and users can name the models they want in the query. The lead is not only a manager: it does part of the work itself, then pulls everything together into one single answer.
3. The agents talk to each other — that is what makes it different from Ultra
Multiagent sits above Ultra in Perplexity Computer's effort slider, and the difference is how the agents communicate. In Ultra, supporting agents report only to the lead, which then assembles the answer. In Multiagent, agents talk to each other directly.
That direct conversation is the whole point. One agent can use evidence another agent found, question an assumption, or spot a discrepancy and route it for further investigation — before the lead writes the final answer. Perplexity presents this as a way to catch mistakes a single model would miss.
4. It is best for large jobs with distinct parts worth cross-checking
Multiagent is designed for large tasks that split into distinct parts and benefit from parallel analysis and cross-checking. Perplexity's own examples are firmly in professional territory: multi-scenario earnings forecasts, annual-report reviews that hunt for numerical inconsistencies, M&A data-room due diligence, and preparation of complex tax filings.
Notice what those examples have in common. Each is a big document-heavy job with parts that can be worked on at the same time, and each punishes a small error — a wrong number in a report, a missed inconsistency in a filing.
Perplexity also says when to use something else. If you want independent perspectives on the same question — different models answering separately, so you can compare their views — use Model Council instead. Multiagent is for one team working together toward one answer, not a panel of separate opinions.
5. What it costs, and what its limits are
Multiagent is available to all Perplexity Max subscribers on the web, selected in Computer's effort slider. Max costs $167 per month billed annually and includes 10,000 Computer credits per month. It is web only at launch.
Two limits are worth knowing before you try it. First, Perplexity has published no benchmarks, accuracy or speed figures for Multiagent, so there is no independent measure of how much the cross-checking improves results. Second, the quality of the answer still depends on the sources and files given to the team: agents can only check the material they have.
Key takeaways
- Perplexity launched Multiagent on 17 November 2026 as a new effort level above Ultra in Perplexity Computer.
- A lead agent running Claude Fable 5.2 plans the task and divides it among up to 16 supporting agents on Claude Opus 5.5, GPT 6.1 Sol and Gemini 4 Carbon.
- Unlike Ultra, where agents report only to the lead, Multiagent agents talk to each other directly, sharing evidence and checking each other's work.
- It is best for large, document-heavy jobs such as earnings forecasts, annual-report reviews, M&A due diligence and complex tax filings.
- Use Model Council instead when you want independent perspectives on the same question rather than one team answer.
- Multiagent is available to Perplexity Max subscribers on the web only; Max costs $167 per month billed annually, including 10,000 Computer credits per month. Perplexity has published no benchmarks for it.
Conclusion
Multiagent is a straightforward idea taken seriously: if one AI can make a mistake, give the job to a team and let the team argue about it first. The model mix is strong and the cross-checking is genuinely different from Ultra. The open question is how much it helps in practice, because Perplexity has published no benchmarks — so the honest way to judge it is to try it on a job where you can check the answer yourself.
FAQ
What is Perplexity Multiagent?
Multiagent is a new effort level above Ultra in Perplexity Computer, launched on 17 November 2026. A lead agent plans the task and divides it among up to 16 supporting agents on different frontier models, which check each other's work before the lead writes one single answer.
How is Multiagent different from Ultra?
In Ultra, supporting agents report only to the lead. In Multiagent, agents talk to each other directly — one agent can use evidence another found, question assumptions, and spot discrepancies before the final answer is written.
How much does Perplexity Multiagent cost?
Multiagent is available to all Perplexity Max subscribers on the web, selected in Computer's effort slider. Max costs $167 per month billed annually and includes 10,000 Computer credits per month. Perplexity has published no benchmarks or accuracy figures for Multiagent.
Sources
- Perplexity — Introducing Computer's Multiagent effort level
- MarkTechPost — Perplexity Computer Adds Multiagent: Up to 16 AI Agents That Cross-Check Work
- Unite.AI — Perplexity Launches Multiagent to Coordinate Frontier AI Models
- Daily Tech & AI — Devin Bots vs Dots vs Duet: AI Coding Managers
- Daily Tech & AI — AI Tools Collection
Researched with AI assistance and reviewed for accuracy.
Follow on WhatsApp
Follow Daily Tech & AI on WhatsApp for one tech and AI update every day: Daily Tech & AI WhatsApp channel.
Comments
Post a Comment