To create this report, Perplexity said Deep Research “iteratively searches, reads documents, and reasons about what to do next, refining its research plan as it learns more about the subject areas,” supposedly “similar to how a human might research a new topic.”
The company also highlighted its performance on Humanity’s Last Exam, an AI benchmarking test with expert-level questions in a variety of academic fields. Perplexity said its Deep Research tool scored 21.1% on the test, easily beating most other models, such as Gemini Thinking (6.2%), Grok-2 (3.8%), and OpenAI’s GPT-4o (3.3%) — but not quite matching OpenAI’s Deep Research (26.6%).