Last week, Perplexity unveiled its new Deep Research feature, designed to enhance the analysis of intricate questions and provide detailed reports through thorough reasoning on various topics. This capability has been integrated into the model picker and is accessible to all users. The company asserts that its response generation is quicker than that of OpenAI’s comparable tool, boasting an average output time of just three minutes. Additionally, Perplexity recently launched an AI assistant for Android, equipped with computer vision and web search functionalities.
Perplexity Unveils Deep Research Feature
Following in the footsteps of Google and OpenAI, Perplexity has emerged as the latest AI entity from Silicon Valley to introduce a Deep Research tool. According to a detailed description on the company’s blog, the feature likely employs a test time compute expansion strategy. This allows the AI model to extend its analysis time on a query, enabling it to investigate the issue more thoroughly, reassess its responses, and explore alternative theories.
The Deep Research feature is being made available to all users, with Pro subscribers enjoying unlimited access to its queries, while free-tier users will have a limited daily allowance. Currently, this feature can only be utilized on Perplexity’s web platform, although the company has plans for it to launch soon on iOS, Android, and Mac applications.
To activate Deep Research, users can select the appropriate mode from a drop-down menu located under the text box. Once activated, the feature allows the AI to commence responding to queries using the Deep Research capabilities. Perplexity claims that it takes roughly three minutes to deliver a response, significantly quicker than the five to 30 minutes required by OpenAI’s tool.
This tool harnesses search and coding functionalities to perform iterative research, analyzing documents and modifying its strategy based on the insights it gathers. After meticulously reviewing the source materials, it organizes the information into a cohesive report. Users have the option to export this report as a PDF or a document or share it as a Perplexity Page with others.
In a noteworthy internal assessment, the Deep Research feature reportedly achieved a score of 21.1 percent on Humanity’s Last Exam, outperforming Gemini Thinking, OpenAI’s o1 and o3-mini models, and DeepSeek-R1.
