Agents on different models split a single task
Perplexity said on its official X account on Nov. 17 (1:28 a.m. Nov. 18 Japan time) that it had begun rolling out "Multiagent," a new "effort level" in "Computer" and "teammates," its AI agents that work on users' behalf. A lead agent splits tasks across a team of agents using different frontier models (the most advanced large AI models). According to the company, the agents share findings with one another as they work, and the lead agent finally combines everything into one answer.
We’re rolling out Multiagent, a new effort level in Computer and teammates. A lead agent splits tasks across a team of agents using different frontier models. As they work, agents share findings with one another, with the lead agent combining everything into one answer.
Teammates, agents that keep working in set roles, were added to Computer on Nov. 5 (Nov. 6 Japan time). Multiagent is being added to both Computer and teammates.
Performance first, with heavy credit consumption
The company described Multiagent as a setting optimized for performance. Saying it is credit-intensive, the company asked users to use it only for their most important tasks. It cited forecasting, high-density problem solving and consequential decision making as the uses it is best suited for.
The company has not disclosed how many credits a single task uses or how much more that is than other effort levels. It also has not stated any cap on the number of uses.
In the company's testing, mixed teams beat single models and copies
Perplexity published results it measured on the benchmark (a test of performance) "Terminal-Bench 4.1." According to the company, it ran five frontier models alone, as two and four copies of the same model, and as the lead of a mixed team of four models. In Multiagent, each mixed team outperformed both the lead model alone and multiple copies of the same model, the company said.
The highest score, 0.727, came from the mixed team led by Gemini 4 Carbon.
On Terminal-Bench 4.1, we ran five frontier models alone, as two and four copies, and leading a mixed-model team of four. In Multiagent, each mixed-model team outperformed both the lead model and multiple copies. The mixed team led by Gemini 4 Carbon scored highest, at 0.727.
These are Perplexity's own measurements, and no third-party verification has been confirmed. The text of the post does not name the five models, the models placed in the mixed teams or the scores of other combinations. ORIENS has not checked the contents of the chart attached to the post.
Launching on the web for Max and Enterprise Max
It is available to "Perplexity Max" and "Enterprise Max" subscribers, on the web. For the mobile and desktop apps, the company said support is "coming soon."
The post does not disclose the monthly price of Max or whether using Multiagent costs extra. It does not mention plans other than Max and Enterprise Max, such as Pro. Whether it can be used from Japan and which languages it supports also could not be confirmed.






