[Perplexity AI Integrates Deep Research: Success Rate for Complex Tasks Doubles] According to monitoring by Beating Insights, Perplexity has announced the integration of Deep Research as a native capability within its AI agent system, Computer, now available to Pro and Max subscription users. The Computer version of Deep Research is built on the new Search as Code architecture and Agent Search SDK. During operation, the model autonomously writes code to construct search workflows, executing thousands of retrieval steps in parallel to deliver customized searches for each specific query. In terms of workflow support, users can import local files and applications to provide internal context and cross-reference real-time web data. The Computer version of Deep Research breaks down complex problems into multiple subtasks and orchestrates over 20 cutting-edge models to collaborate on processing. For instance, the system assigns legal reasoning models to review contracts, uses data models to analyze Excel discrepancies, and calls writing models to generate final reports. Within Computer, users can complete the entire process from research to deliverable creation, with the system directly returning research reports, presentations, and dashboards. Benchmark results show that the integration of Deep Research into Computer outperforms the previous version across multiple evaluations. The new model improved accuracy in OpenAI's BrowseComp test from 40.7% to 83.8%. In Humanity's Last Exam, jointly launched by the Center for AI Safety (CAIS) and Scale AI, accuracy increased from 36.4% to 50.5%. For Google DeepMind's DeepSearchQA test, accuracy rose from 81.9% to 85.0%. [Original Link]