Perplexity's new feature pits three frontier models against each other

If you do serious research with AI, you probably know the drill: opening three browser tabs, pasting the exact same prompt into ChatGPT, Claude, and Gemini, and squinting at your screen to see who got it right.

Perplexity wants to automate that whole headache for you.

Perplexity rolled out Model Council, a new multi-model research feature designed to query multiple frontier models simultaneously and synthesize them into a single, cohesive answer.

Three frontier models walk into a prompt

The way Model Council works is essentially like convening a panel of experts.

When you trigger the feature on Perplexity's web interface, your prompt runs across three top-tier models at the same time, such as Claude Opus 4.6, GPT 5.2, and Gemini 3.0.

Once the individual responses are generated, a synthesizer model steps in to review the outputs. It resolves contradictions where possible and produces a unified report.

Crucially, the resulting answer highlights where the models agree and points out where their reasoning diverges.

Every frontier model has blind spots, whether that means leaning toward certain biases or hallucinating details with total confidence. Forcing them to cross-examine each other makes those gaps immediately visible.

When to call in the council

Perplexity says the tool is built for high-stakes workflows where accuracy and multiple perspectives genuinely matter.

One obvious use case is investment research, where leaning on a single biased model could lead to costly mistakes.

It is also aimed at complex strategic decisions, like evaluating career moves or major purchases, where different reasoning styles add value. It can even be used for brainstorming trips or content ideas by drawing from varied model strengths.

As of now, Model Council is available on the web exclusively for Perplexity Max subscribers, with mobile app support slated to arrive soon.

Why it matters

For power users, Model Council eliminates the friction of manual cross-verification.

We all know LLMs hallucinate, but almost nobody has the patience to manually copy and paste prompts between three different apps all day. Showing convergence and divergence in one view gives you instant confidence when models agree, and a clear signal to double-check sources when they fight.

That said, the barrier is real. Gating it behind the premium Perplexity Max tier means everyday free users will not touch it anytime soon, and running four model calls per query undoubtedly introduces latency and high compute overhead.

Strategically, we think this is a brilliant flex of Perplexity's aggregator advantage.

Perplexity does not need to build the best foundation model in the world when it can simply orchestrate everyone else's. Proprietary model makers like OpenAI or Anthropic would never willingly serve you their competitors' outputs side-by-side, which leaves multi-model orchestration as an open lane for Perplexity to dominate.