AI Aulis

AI NEWS IN ENGLISH · 19 November 2026

The most important AI news every morning. AI writes and checks the stories against the original sources – check important facts in the sources.

38 stories · next issue tomorrow morning · How the paper is made →

← Issue 19 November 2026Paper

In Perplexity Computer’s Multiagent level, a lead agent splits work among up to 16 agents

AI-generated contentWritten by AI

perplexity.ai, aitimes.com · 2 sources · Sources published: 17–18 November 2026

Perplexity says the agents share findings directly with each other, check each other’s work and run on models from Anthropic, OpenAI and Google. The level is available to Perplexity Max subscribers on the web.

On 17 November, Perplexity launched Multiagent, a new reasoning effort level for its Computer service. Computer is a platform for AI agents. Its agents research complex questions and carry out multi-step work such as analysing data and building apps. The reasoning effort level lets users choose how much work Computer puts into a task. The new level sits above Ultra.

The options differ in how they organise the work. In Ultra, a lead agent delegates subtasks to subagents, which report only to the lead agent. In Multiagent, agents share findings directly with each other: one can use evidence another has found, challenge an assumption or flag a contradiction for further investigation. In the third option, Model Council, several models research the same question independently. The service shows where the models agree and where they disagree.

Perplexity wrote on its blog that Claude Fable 5.2 plans the work and divides it among agents that run Claude Opus 5.5, GPT 6.1 Sol and Gemini 4 Carbon by default. The lead agent can bring in up to 16 agents and do part of the work itself. Users can also name the models they want in the prompt.

The agents edit shared files. In the company’s example, one agent updates the revenue assumptions in a financial model and notifies another, which uses the figures in a valuation.

The announcement gives no measurements of whether cross-checking makes answers more accurate. South Korea’s AI Times summarised the same announcement, so its report does not independently confirm the company’s description. In some of its use cases, the company presents the output as a basis for a professional’s work. In an annual report, the agents flag conflicting figures for an auditor to assess, and in tax planning they prepare questions for a tax adviser.

Users can see the agents and their models while the work runs. Stopping the conversation thread stops the whole team. Multiagent is available to all Perplexity Max subscribers in the web version. In an X post quoted by AI Times, Perplexity warned that the level uses a lot of credits. The company advised using it only for the most important tasks.

Previous coverage

Report an error in this story