Skip to main content

Perplexity launches Multiagent for complex Computer tasks

18 NOVEMBER 2026·2 MIN READ·3 SOURCES

Perplexity introduced Multiagent, a new effort level for Computer’s most complex work. The feature coordinates agents running different frontier models to work in parallel and cross-check findings.

Key takeaways · 4

  • 01

    Multiagent is designed for tasks with distinct parts that benefit from parallel analysis and cross-checking.

  • 02

    The default supporting team is Claude Opus 5.5, GPT 6.1 Sol and Gemini 4 Carbon, coordinated by Claude Fable 5.2.

  • 03

    Users can see which agents and models are running and guide the work through the lead agent.

  • 04

    Perplexity calls Multiagent credit-intensive, but its launch thread did not publish a price or credit figure.

A lead agent coordinates the work

Perplexity introduced Multiagent as a new effort level above Ultra for Computer’s most complex work.[1] The lead agent can assign work to as many as 16 agents while handling part of the task itself.[1] Perplexity says the agents run different frontier models, work in parallel, share findings directly and check one another’s work.[1] The lead then combines their findings into a single response with the requested deliverables.[1] Perplexity describes the feature as suited to work with distinct parts that benefits from parallel analysis and cross-checking.[1]

The model lineup and handoffs

Perplexity says Claude Fable 5.2 plans the task and divides it among agents running Claude Opus 5.5, GPT 6.1 Sol and Gemini 4 Carbon.[1] Those three models are the default supporting team.[1] Claude Fable 5.2 and GPT 6 Astra can also join as supporting agents when requested or when a task component calls for them.[1] That setup differs from Ultra agents, which Perplexity says report only to the lead.[1] For a valuation task, Perplexity gives the example of one agent updating revenue assumptions and notifying another agent to use them in its analysis.[1]

Visibility, control and availability

Users can see which agents and models are running during a task, according to Perplexity.[1] They can guide the work through the lead agent without messaging each supporting agent separately.[1] Perplexity says that when changes affect agents, the lead tells them on a follow-up and the team continues its existing work.[1] Multiagent is available to all Perplexity Max subscribers on the web; ORIENS also reports web availability for Enterprise Max subscribers.[1][2] Perplexity describes the feature as credit-intensive and recommends reserving it for important tasks, while CellCog reports that no price or credit figure appeared in the launch thread.[3]

Tests show gains, with reporting limits

CellCog reports that Perplexity tested five frontier models alone, as two and four copies, and leading mixed teams of four.[3] In Terminal-Bench 4.1 results, every mixed team beat four copies of its lead model by 0.008 to 0.125, according to CellCog.[3] The Gemini 4 Carbon-led team scored 0.727 in a run costing $10.31; CellCog reports that the top mixed team cost $10.31 per run, compared with about $13 for four copies of its lead model.[3] CellCog also reports mixed-team gains on WANDR, but says the launch thread gives no run counts per score or spread between runs.[3] The published results therefore do not show how much scores varied across repeated runs.

Teams considering multi-model workflows can assess whether parallel analysis and cross-checking fit their tasks, while keeping an eye on usage costs. The absence of a published credit figure and run-to-run reporting makes it harder to estimate expense and judge how stable the reported benchmark gains are.

Why it matters
For teams

AI training for your team

kju.ai is built for teams: daily sessions, team analytics, and rollout support for your organization.

How this developed

  1. 18 November 2026

    Perplexity launches Multiagent for complex Computer tasks

Sources

AI fluency, one session a day, built for your work.