Model Council at a glance

Model Council is a Perplexity search and research feature that runs one question across multiple AI models in parallel and produces a unified synthesis.

The synthesis is organized around convergence, disagreement and findings that only one model surfaced. Each contributing response remains visible so the user can inspect what the synthesis compressed or omitted.

On the web and in the iOS app, a council always uses three models. Users can swap eligible models in the picker and toggle deeper Thinking where a model supports it.

As of this review, Perplexity lists the default web lineup as GPT-5.6 Sol with Thinking, Claude Opus 5 with Thinking and Gemini 3.1 Pro with Thinking. The company warns that the exact model list changes as providers release and retire models.

In Perplexity Computer, Model Council works differently: users choose two to eight models and an analysis depth, receive the synthesis plus each model's full report as a file, and can ask Computer to package results into a PDF, deck or spreadsheet.

Web and iOS access is included only with Max and Enterprise Max and has a monthly query cap that Perplexity does not publish as a fixed number on the help page. Computer access is available on every paid plan but consumes credits for each model and depth selected.

The iOS search mode requires app version 26.16.0 or later. Perplexity says the standalone mode is not yet available in Android or desktop apps, although Computer provides another route on paid plans.

Model diversity is useful for finding different assumptions, but the models may rely on the same popular sources, inherit similar training data or repeat the same false claim. Three agreeing outputs are not equivalent to three independent experts or primary sources.

The synthesizer is another model and can flatten minority evidence, resolve a conflict incorrectly or make the final answer sound more certain than the underlying reports justify. Review the individual outputs when the decision matters.

Consumer Free, Pro and Max accounts have AI data retention enabled by default and must turn the setting off prospectively to opt out of model-training use. Perplexity says Enterprise data is never used for training and third-party providers operate under zero-training and zero-retention agreements.

For consequential research, ask each model to state its assumptions and confidence, inspect the cited primary evidence, document unresolved disagreement and have an accountable domain expert make the final decision.