Multi-Model Orchestration with 'Computer'

Perplexity has integrated its Deep Research capability into 'Computer,' a cloud-based orchestration system launched in February 2026. This system acts as a model-agnostic coordinator, utilizing Opus 4.6 as its primary reasoning engine while delegating specialized subtasks to other frontier models (e.g., using Gemini for specific research tasks). By breaking complex user prompts into granular subtasks, the system achieves higher accuracy and deeper analysis than previous iterations.

Search as Code and Agentic Workflows

The core innovation is 'Search as Code,' where the system dynamically generates code to execute search strategies. Instead of relying on a static, fixed pipeline, the model writes scripts that run thousands of retrieval steps in parallel. These scripts operate within a sandbox, utilizing the Agent Search SDK to perform filtering, deduplication, and reranking. This approach allows the system to branch, compare, and refine its research path in real-time. The system also supports multi-modal context, allowing users to upload PDFs or spreadsheets to be cross-referenced against live web data and premium sources like PitchBook and CB Insights.

Performance and Developer Access

Perplexity reports significant performance gains in agentic browsing, with the 'BrowseComp' benchmark score jumping from 40.7% to 83.8%. Developers can access this stack via the pay-as-you-go Agent API, which supports a deep-research preset. The API is compatible with OpenAI SDK standards, allowing for integration into custom applications. The system is designed to produce work-ready deliverables, including reports, decks, and dashboards, with a preview-and-approve workflow for any changes made to user-provided files.