Perplexity Enhances Deep Research with Multi-Model Orchestration in Computer

Perplexity has significantly upgraded its Deep Research feature by integrating it into Computer, its advanced multi-model orchestration system. This integration allows Deep Research to break down complex questions into subtasks and route them across more than 20 frontier AI models, ultimately producing detailed reports, presentations (decks), and dashboards. The move is designed to enhance the accuracy, analytical depth, and citation quality of the generated outputs.

The upgraded Deep Research is now a core component of Perplexity Computer, which initially launched in late February 2026. Computer acts as a cloud-based system capable of coordinating up to 20 AI models within a single workflow. Its design is model-agnostic, with Opus 4.6 serving as its primary reasoning engine. Specialized tasks are handled by sub-agents, such as using Gemini for specific deep research requirements.

Key facts

Feature	Description
Integration	Deep Research now lives within Perplexity Computer
Models Used	Routes subtasks across 20+ frontier AI models
Output	Generates work-ready reports, decks, and dashboards
Core Engine	Opus 4.6 serves as the primary reasoning engine for Computer

Enhanced Research Capabilities

The new Deep Research in Computer is built upon two foundational components: the Agent Search SDK and Search as Code. When presented with a complex query, the system automatically formulates a research plan. It then proceeds to locate and gather information from primary sources across hundreds of websites, citing every claim made in the generated output. The system achieves this by writing code that orchestrates the search process itself. This code executes thousands of retrieval steps in parallel, adapting each step to the specific requirements of the question.

This code-driven search operates within a sandboxed environment and utilizes Perplexity’s Agentic Search SDK. The SDK provides essential search primitives, including filtering, deduplication, and reranking capabilities. This approach is a departure from fixed pipelines, which follow a predetermined set of steps regardless of the query. The flexibility of code-driven search allows the system to branch, compare, and refine its search strategy as it learns and progresses.

Accessibility for Developers and Users

Search as Code is being rolled out through both the Computer platform and the Agent API, enabling developers to programmatically access the same agentic search stack. Perplexity Max users can access Deep Research in Computer as a consumer feature, while developers can utilize the pay-as-you-go Agent API. The official SDK includes a deep-research preset, facilitating easy integration for developers.

Beyond web searches, Computer can also process user-provided files, such as PDFs or spreadsheets, integrating this internal context with live web data. This allows for cross-referencing user data with external sources like census data or Statista.

Performance Improvements

Perplexity has released comparative performance data, highlighting the gains achieved by the Computer version of Deep Research over its legacy counterpart. The most significant improvements were observed in agentic browsing tasks, where the system must navigate numerous web pages to find elusive information. In BrowseComp tests, which assess an agent’s ability to locate hard-to-find information through browsing, the performance jumped from 40.7% to 83.8%. While DeepSearchQA, which covers expert questions across various academic subjects, already performed well, it also saw a positive, albeit smaller, gain.

Specialized Subtask Routing

A key advantage of Computer’s multi-model approach is its ability to route each subtask to the AI model best suited for it. For instance, a legal reasoning model would handle contract reviews, a data model would manage spreadsheet variance checks, and a writing model would craft the final draft. This specialized allocation ensures optimal efficiency and accuracy for each component of the research process. The answers are further bolstered by premium data sources, including PitchBook and CB Insights, with legal data currently available in preview.

Source: MarkTechPost: https://www.marktechpost.com/2026/06/11/perplexity-moves-deep-research-into-computer-routing-research-subtasks-across-20-frontier-models-for-reports-decks-and-dashboards/