Perplexity has added Multiagent to Computer, its agent platform for analysis and multistep work. Perplexity Computer Multiagent is a brand new effort degree that sits above Extremely. A lead agent runs a workforce of brokers on totally different frontier fashions. These brokers work in parallel, share findings straight and examine one another’s work. For builders and analysts, the change is structural. Supporting brokers not report solely to a single lead.
TL;DR
- Dimension: As much as 16 brokers per job, plus the lead. Context size: Not disclosed.
- Runs on: Perplexity’s cloud. Net solely, Perplexity Max plan. No native or quantized choice.
- Efficiency: Not benchmarked. Perplexity printed no accuracy or pace numbers.
- Greatest: 16-agent cap with direct agent-to-agent checks throughout 3 default mannequin households (Claude, GPT, Gemini).
What’s Perplexity Pc Multiagent?
Multiagent is the very best effort degree in Perplexity Pc. It coordinates AI fashions to research advanced duties and cross-check their work, per Perplexity’s announcement.
It builds on effort controls launched on September 17, 2026. That launch added 4 settings: Mild, Customary, Excessive and Extremely. Every setting picks the orchestrator mannequin and its reasoning depth. Decrease settings use inexpensive fashions.
How does Multiagent work?
The lead agent plans the duty and assigns elements to supporting brokers. It might probably assign as much as 16 brokers and likewise does a part of the work itself.
In Perplexity’s instance, Claude Fable 5.2 plans the duty. It splits work amongst brokers working Claude Opus 5.5, GPT 6.1 Sol and Gemini 4 Carbon. These 3 fashions type the default workforce. Fable 5.2 and GPT 6 Astra be part of as supporting brokers solely when requested or wanted.
Key mechanics from the launch post:
- Direct sharing: Brokers use proof discovered by others, query assumptions and flag discrepancies.
- Shared recordsdata: Brokers coordinate edits to shared recordsdata. One agent can replace income assumptions and notify one other to reuse them in a valuation.
- Single level of contact: Customers discuss solely to the lead. The lead merges findings into one response.
- Comply with-ups: The lead tells affected brokers what modified. The workforce continues its current work.
- Management: Customers see which brokers and fashions are working. Stopping the thread stops the entire workforce.
- Mannequin alternative: Customers can title particular fashions of their question.
Multiagent vs Extremely vs Mannequin Council
Perplexity now provides 3 methods to place a number of fashions on 1 drawback:
- Extremely: One lead delegates to supporting brokers. These brokers report solely to the lead.
- Multiagent: The identical lead mannequin as Extremely coordinates brokers. They discuss to one another and edit shared deliverables.
- Mannequin Council: A number of fashions reply the identical query independently. Pc then exhibits the place they agree and disagree. Model Council came to Computer in July 2026.
Perplexity positions Multiagent for big, interconnected work that may run in parallel. Its examples embrace deal information room diligence, multi-scenario forecasts, annual report overview, competing-hypothesis analysis and year-end tax preparation.
(perform(){var f=doc.getElementById(‘mtp-pplx-ma-frame’);window.addEventListener(‘message’,perform(e){if(e.information&&e.information.mtpEmbed===’pplx-ma’&&e.supply===f.contentWindow){f.fashion.top=e.information.top+’px’;}});})();
How does it examine with different multi-agent merchandise?
| Characteristic | Perplexity Pc Multiagent | Kimi Agent Swarm | Manus Extensive Analysis | SuperGrok Heavy |
|---|---|---|---|---|
| Max parallel brokers | 16 (plus lead) | As much as 300 sub-agents | “Hundreds”; examined as much as 250 objects | Not disclosed |
| Agent-to-agent communication | Sure, direct, plus shared recordsdata | Not disclosed | Not disclosed (most important agent assembles outcomes) | Brokers examine work (per xAI launch claims) |
| Fashions used | Blended distributors: Claude Opus 5.5, GPT 6.1 Sol, Gemini 4 Carbon (default) | Kimi K3 (K3 Swarm) | Not disclosed | Grok 4 Heavy |
| Printed efficiency information | None | 4.5x sooner than single-agent sequential (K2.6) | None | Not disclosed for present model |
| Platforms | Net solely | Net and cell app | Net | Not disclosed |
| Plan / value | Perplexity Max, $167/month billed annually | Not disclosed | Not disclosed | $300/month at launch |
The clearest distinction is mannequin variety. Perplexity mixes Anthropic, OpenAI and Google fashions in 1 workforce. Kimi and Grok run their very own mannequin households. Kimi trades depth of debate for scale, with as much as 300 sub-agents.
Deployment and availability
Multiagent is accessible to all Perplexity Max subscribers. It runs on the net solely. Customers choose Multiagent in Pc’s effort slider when beginning a brand new job.
Perplexity’s pricing page lists Max at $167 per 30 days billed yearly. Max contains 10,000 Pc credit per 30 days. The launch publish doesn’t state credit score price per Multiagent job.
Key Takeaways
- Multiagent is a brand new Pc effort degree above Extremely.
- A lead runs as much as 16 brokers that share findings straight.
- Default workforce: Claude Opus 5.5, GPT 6.1 Sol and Gemini 4 Carbon.
- Max-only, web-only, with no printed benchmarks.
- Greatest match: information room diligence and multi-scenario forecasting.
Take a look at the official announcement, the effort controls post and Perplexity pricing. You possibly can examine the deployment choices and specs right here via our GPULeaderboard.com chrome extension.
Additionally, be at liberty to comply with us on Twitter and don’t neglect to affix our 150k+ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well.
Must accomplice with us for selling your GitHub Repo OR Hugging Face Web page OR Product Launch OR Webinar and so forth.? Connect with us
The publish Perplexity Computer Adds Multiagent: Up to 16 AI Agents That Cross-Check Work appeared first on MarkTechPost.

