Perplexity has added a Multiagent effort level to Computer, its agent platform, in which a lead agent coordinates a team of agents running on different frontier models. The company announced it on 17 November 2026. It sits above the existing Ultra setting and is available to all Perplexity Max subscribers on the web.
Quick Take
- What: a new effort level above Ultra where agents work in parallel, share findings directly and check each other’s work.
- Team size: the lead can assign up to 16 agents while also taking on part of the work itself.
- Default models: Claude Fable 5.2 plans and divides the task; agents run Claude Opus 5.5, GPT 6.1 Sol and Gemini 4 Carbon.
- Optional models: Fable 5.2 and GPT 6 Astra join as supporting agents when a user asks or a task calls for them.
- Availability: Perplexity Max subscribers on the web, selected in Computer’s effort slider.
In plain English
Instead of one AI working through a big task alone, Computer now assigns pieces to several AIs from different companies. They can pass evidence to each other and flag disagreements, and the lead agent merges everything into one answer. You can see which agents are running and which model each uses.
What happened
Perplexity said Multiagent is aimed at complex work such as forecasting, annual report review, deal data room diligence and tax planning. In a forecast, for example, one agent can update revenue assumptions in a financial model and notify another to use those figures in its valuation analysis. Stopping the thread stops the whole team.
Background
In September Perplexity introduced effort controls with four settings, in which one lead agent plans the task and delegates parts to supporting agents that report only to the lead. Perplexity also offers Model Council, where several models investigate the same question independently. Read our earlier report on Perplexity Teammates and the original Perplexity Computer launch breakdown.
How it works
Perplexity distinguishes the three options. Ultra has the lead delegate to agents that report back only to it. Multiagent uses the same lead model but lets agents share findings directly and contribute to shared deliverables. Model Council gives independent perspectives on the same question before combining them. Users can name the models they want in their query.
Why it matters
For professionals in finance, audit, legal and tax, cross-checking between models is meant to surface inconsistent figures before a human reviews them.
For model vendors, the default team mixes Anthropic, OpenAI and Google models, showing agent platforms treating models as interchangeable parts.
For buyers, the feature is limited to the Max tier, so it adds to the premium end of the lineup.
What’s still open
- Pricing or usage limits for Multiagent runs beyond Max access, which Perplexity did not detail.
- Independent evidence that cross-checking between agents improves accuracy over a single agent.
- Availability on mobile and for other plans, as only the web is mentioned.
Sources
- Perplexity, “Introducing Computer’s Multiagent effort level,” 17 November 2026
i10x Perspective
This is a vendor announcement with no benchmark attached, so the claim that agents catch each other’s errors remains untested publicly. It does show where agent products are heading: orchestration across rival models. More in our Agents section and on the Perplexity company page.






