Perplexity announced Multiagent, a high-effort execution tier for its research and workflow platform, Perplexity Computer. Sitting above the existing Ultra tier, Multiagent deploys a primary lead agent to coordinate up to 16 supporting sub-agents running across distinct frontier models from Anthropic, OpenAI, and Google. The sub-agents operate in parallel, sharing research findings directly and cross-checking outputs to perform complex analyses.

In a demonstrated configuration, Claude Fable 5.2 acts as the planning lead, orchestrating tasks across Claude Opus 5.5, GPT 6.1 Sol, and Gemini 4 Carbon. The multi-model approach targets intricate financial, legal, and operational workflows such as data room due diligence, multi-scenario forecasting, and annual report audits. The feature is available immediately to Perplexity Max subscribers on the web.

By pooling heterogeneous model families into a unified execution framework, Perplexity aims to mitigate single-model biases and hallucinations. The launch underlines a broader industry movement toward agentic multi-model orchestration platforms capable of executing long-horizon enterprise tasks.

Why it matters

  • Demonstrates practical multi-model orchestration, allowing teams to leverage complementary strengths of competing frontier LLMs.

  • Enables automated verification workflows where parallel agents audit each other’s research to reduce hallucination risk.

  • Provides a template for complex enterprise automation in research-heavy sectors like finance, legal, and consulting.

Source: marktechpost.com