Virtual Reality Polout February 3, 2025 0 Comments

OpenAI’s New Deep Research AI Surfs the Web, Writes Papers

ChatGPT can now conduct research and write a report as thoroughly as a research analyst, OpenAI claims.

OpenAI launched a new AI agent capability on Sunday called Deep Research, which infuses ChatGPT with the power to search the web, analyze sources, and determine if those sources are relevant. Based on its findings, Deep Research produces a comprehensive research paper with full citations that can sometimes run longer than 10,000 words.

“It can do complex research tasks that might take a person anywhere from 30 minutes to 30 days,” OpenAI’s chief product officer Kevin Weil stated at an event in Washington showcasing the technology last week. Deep Research can get the same tasks done in five to 30 minutes, he said.

While AI chatbots provide conversational responses to user queries, AI agents go a step further by completing multi-step tasks on their own. Deep Research is an AI agent that can surf the Internet and synthesize information from various sources, and it is the second AI agent from OpenAI this year. OpenAI introduced Operator last month, an AI agent that can shop for groceries online or make a restaurant reservation.

Deep Research has already beat out the competition, including DeepSeek’s R1, xAI’s Grok 2, and Google’s Gemini Thinking, on a new evaluation called “Humanity’s Last Exam” released last month by researchers at the Center for AI Safety and Scale AI.

Researchers claim that the 3,000-question exam is the most difficult test ever given to AI systems. While DeepSeek’s R1 scored 9.1% on the test, Google’s Gemini Thinking scored 6.2%, and Grok 2 scored 3.8%, Deep Research scored 26.6%.

How does Deep Research work?

In a demo introduction to Deep Research released on YouTube on Sunday, OpenAI product project manager Neel Ajjarapu demonstrated how Deep Research was accessible as a button within the search bar of ChatGPT.

Accessing Deep Research is as simple as pressing a button on the homepage of ChatGPT.

Ajjarapu wanted Deep Research to look into whether OpenAI should develop a new language translation app and typed in a specific query: “Help me find iOS and Android adoption rates, % who want to learn another language, and change in mobile penetration, over the past 5 years, for top 10 developed and top 10 developing countries by GDP. Lay this info out in a formatted report, a table on metrics, and include recommendations on markets to target for a new translation app from ChatGPT, focusing on markets ChatGPT could better expand to.”

Deep Research responded to Ajjarapu’s prompt with a series of questions, including if he wanted overall iOS and Android adoption rates, or specific categories, like device shipments.

After Ajjarapu answered the clarifying questions, Deep Research was off to work. The AI agent kept Ajjarapu updated on what it was doing in ChatGPT, like searching for the top 10 countries by GDP.

^{While DeepResearch goes to work, it keeps users updated on what it’s doing in the sidebar to the right. Credit: OpenAI}

Meanwhile, Deep Research was surfing the web on its own, understanding text, images, and PDFs, and synthesizing information. Ajjarapu said that Deep Research could be used for market research, academic research across departments, and at work.

In 11 minutes, Deep Research looked through 29 websites in-depth to create a research report complete with tables, bullet points, and cited sources. This report would have taken “hours” to put together, Ajjarapu stated.

Deep Research is now embedded within ChatGPT for OpenAI Pro users, with plans to roll it out to paying Plus and Team users next.