Perplexity Launches Model Council to Run Queries Across Multiple AI Models

TL;DR: Perplexity has launched Model Council, a feature that runs queries across three AI models simultaneously and uses a synthesiser to resolve conflicts between their outputs. The tool highlights where models agree and where they diverge, giving users a clearer picture of confidence levels across different responses.

AI model performance increasingly varies by task. A model that excels at coding may produce mediocre research summaries. One that handles creative writing well might stumble on factual analysis. Perplexity’s Model Council addresses this by running the same query across multiple models and presenting a synthesised result.

How It Works

When a user selects Model Council, their query runs across three models — currently Claude Opus 4.6, GPT 5.2, and Gemini 3.0. Each model produces its own response independently. A synthesiser model then reviews all three outputs, identifies areas of agreement, flags contradictions, and produces a consolidated answer that makes the level of consensus visible.

The logic is straightforward: when all three models converge on the same answer, users can move forward with greater confidence. When they disagree, users know to dig deeper before acting on any single response. Every AI model has blind spots, and cross-referencing reduces the chance of relying on one model’s particular weakness.

Best Use Cases

Perplexity recommends Model Council for tasks where accuracy matters more than speed: investment research, complex business decisions, creative brainstorming where diverse perspectives add value, and verification of factual claims. These are areas where a single model’s confident-sounding but incorrect answer could lead to poor decisions.

Looking Forward

Model Council is available now for Perplexity Max subscribers on web, with mobile support coming soon. The feature reflects a broader shift in how people interact with AI — moving from reliance on a single model towards multi-model approaches that treat each system’s output as one input among several. As the performance gap between models narrows on straightforward tasks but widens on specialist ones, tools that aggregate and compare outputs across models are likely to become more common. The question is whether users will take the time to examine disagreements or simply default to the synthesised summary.