Products & Tools

Perplexity Launches Multiagent Mode to Coordinate Frontier AI Models

Perplexity's November 17 Multiagent setting lets a Claude Fable 5.2 lead coordinate up to 16 agents across Claude, GPT and Gemini models that share findings directly.

Perplexity Launches Multiagent to Coordinate Frontier AI Models
Perplexity Launches Multiagent to Coordinate Frontier AI ModelsAlexandra_Koch / pixabay
By Marcus Bennett5 min read

Why it matters

  • Perplexity launched Multiagent on November 17, 2026, allowing up to 16 supporting agents coordinated by one lead.
  • Claude Fable 5.2 leads the default team of Claude Opus 5.5, GPT 6.1 Sol and Gemini 4 Carbon.
  • Multiagent sits above Ultra and is available to all Perplexity Max subscribers on the web.
  • The setting extends effort controls introduced in September 2026: Light, Standard, High and Ultra.
  • Unlike Ultra, Multiagent agents exchange findings directly rather than reporting only to the lead.

Perplexity introduced Multiagent on November 17, 2026, a new effort setting for its Computer agent platform that lets a single lead model coordinate up to 16 supporting agents running across several frontier AI systems to analyze complex tasks, share evidence and cross-check one another before delivering a single response.

The tier sits above Ultra, the previous top setting, and it changes how information moves inside Computer's agent teams. Perplexity says that, unlike Ultra's supporting agents, which report only to the lead, Multiagent's agents can exchange findings directly, question assumptions and flag discrepancies for further investigation. The design is intended to surface conflicting assumptions before the lead combines the team's work into the requested recommendation or other deliverable.

The launch matters because it pushes multi-model orchestration — the practice of routing one assignment across several competing AI systems rather than relying on a single model — into a consumer-subscription product. Perplexity's stated use cases, annual-report review and deal data-room diligence, target professional workflows where errors carry audit and legal consequences.

Which models run the team?

Claude Fable 5.2 serves as the lead, planning the work and dividing it among agents running Claude Opus 5.5, GPT 6.1 Sol and Gemini 4 Carbon. Those three models form the default supporting team. Fable 5.2 and GPT 6 Astra can also join as supporting agents, but only when the user requests them or a particular task component calls for them.

The lead model does more than delegate. Perplexity says the lead remains the conversational point of contact throughout a task, and it can also take on part of the assignment itself. Stopping the thread stops the entire team.

What does peer-to-peer agent communication change?

The structural difference from earlier tiers is the direction of information flow. In Ultra, supporting agents report up a single chain of command. In Multiagent, agents on the same assignment talk to each other: they exchange findings, question each other's assumptions and flag discrepancies that warrant further investigation.

Perplexity frames this as a defect-detection mechanism. When several agents examine distinct but connected parts of a document set, direct exchange lets the team identify conflicting figures or assumptions before the lead assembles the final deliverable, rather than after.

The company's annual-report example illustrates the intended division of responsibility: separate agents examine financial statements, footnotes and management commentary, then compare their findings to flag inconsistent figures or disclosures for an auditor to review. The agents do the cross-checking; the human professional remains the final arbiter. Perplexity describes Multiagent's coordination as preparation for professional scrutiny, with one lead still responsible for assembling the final response.

A second example involves deal diligence on a data room. Separate agents review financial records, contracts and operating documents in parallel, then connect details across the set — for instance, a customer contract that changes the revenue outlook derived from the financial records.

How do users control the team?

Users select Multiagent through Computer's effort slider, the same control that governs the platform's other effort levels, and can name the models they want directly in their query. During a task, users can see which agents are active and which models those agents are running.

The lead stays the single point of contact. Users can redirect the work mid-task without messaging every agent separately, a design choice that keeps the interface conversational even as the underlying team expands to 16 agents. Multiagent is available to all Perplexity Max subscribers on the web.

Can agents edit shared files together?

Yes. Agents can coordinate edits to shared files, and Perplexity gives a concrete financial-modeling example: one agent updates revenue assumptions in a financial model and notifies another to use those figures in its valuation analysis.

Follow-up requests work through the same hierarchy. On a follow-up, the lead tells affected agents what changed, and the team continues its existing work rather than starting over. The continuity applies until the user stops the thread, which halts every agent at once.

Where does Multiagent fit among Perplexity's effort tiers?

Multiagent extends the effort controls Perplexity introduced in September 2026, which offered Light, Standard, High and Ultra settings. Perplexity selected the coordinating model and reasoning level for each setting, while custom controls let users choose a specific model and reasoning level themselves. Lower settings used less expensive models, and credit usage depended on the amount of work Computer performed.

The four September tiers plus Multiagent now form a five-level ladder with distinct intended uses:

  • Light, Standard and High — graduated effort and cost levels using progressively more capable models.
  • Ultra — open-ended problems that require several approaches to be tested in parallel.
  • Multiagent — large assignments with distinct but interconnected parts that can proceed in parallel under a coordinating lead.

Perplexity also positions Model Council separately from both Ultra and Multiagent. Model Council targets judgment calls that benefit from independent perspectives: several models investigate the same question separately, after which Computer combines their findings and shows where they agree and disagree. The distinction is the shape of the task — Model Council applies independent viewpoints to one question, while Multiagent divides the parts of one large assignment among specialists.

Why the tiering signals product direction?

The November 17 release completes a progression Perplexity began two months earlier. The September 2026 controls gave users a model-and-effort tradeoff; Multiagent converts that slider into an organizational chart, with a named lead, a default roster and rules for how agents communicate.

The economics follow the same logic as the earlier tiers. Perplexity disclosed that lower effort settings used less expensive models and that credit usage scaled with the work Computer performed, which implies that a 16-agent assignment across four frontier models will consume more credits than a single-model task. The company did not publish specific pricing figures for Multiagent usage in the announcement.

For subscribers, the practical test will be whether peer-checked agent teams catch the inconsistencies that single-model passes miss — before the auditor does.

Source: unite.ai

Share this article:

More from Marcus Bennett

Marcus Bennett

Show full bio

Senior reporter covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.

718 articles

Related articles

  1. Perplexity launches Teammates: AI agents that own ongoing work, not just chats
  2. Perplexity’s Orchestrator retries failed requests across providers
  3. Anthropic: One AI agent now beats fleets of subagents
  4. AI Agents Move Onto Windows PCs, and Permissions Become the New Battleground
  5. Anthropic scales Claude to 1,000 parallel agents, finds 66 of 70 bugs

« Previous articleNext article »