OpenAI’s ‘deep research’ Might Just Out-Think Google and DeepSeek

OpenAI has launched deep research, a new capability in ChatGPT that independently conducts multi-step research on the internet. The tool can handle complex tasks in a fraction of the time it would take a human researcher.

“Today, we launch deep research, our next agent. This is like a superpower—experts on demand! It can use the internet, conduct complex research and reasoning, and give you back a report. It is really good and can do tasks that would take hours or days and cost hundreds of dollars,” said OpenAI chief Sam Altman in a post on X.

The feature will be available to Pro users starting today, with a limit of up to 100 queries per month. Plus and Team users will gain access next, followed by Enterprise customers.

Deep research couldn’t have been launched at a better time—it came just when DeepSeek’s R1 model was stealing the spotlight. During their live stream and feature demo, a cheeky ChatGPT message read, “Is Deeper Seeker a good name?” OpenAI doesn’t miss a chance, does it?

Deep research uses an optimised version of OpenAI’s upcoming o3 model, which specialises in web browsing and data analysis. According to OpenAI, the tool can search, interpret, and synthesise information from text, images, and PDFs, adjusting its approach based on newly discovered insights.

Users can access deep research by selecting it in the message composer in ChatGPT. The tool allows for file attachments to provide additional context. Once initiated, a sidebar displays a summary of the steps taken and sources used. The results are delivered in five to 30 minutes.

“The model takes a complex question, breaks it down to clarify requirements, searches the internet and files we have provided, and does multi-step reasoning over many minutes to create the answer.  This would be really useful for compiling research reports, shopping advice, travel plans and more,” said VP of OpenAI Srinivas Narayanan.  

New Tool for Consultants and Researchers

The feature is intended for users in finance, science, policy, and engineering, as well as consumers conducting in-depth research on major purchases. OpenAI said that every output includes clear citations and a summary of the reasoning process.

“Deep research independently discovers, reasons about, and consolidates insights from across the web,” OpenAI explained, adding that it was trained using real-world tasks that require browser and Python tool use.

OpenAI’s Isa Fulford introduced deep research and demonstrated how it can make investment research easier. In the live demo, she explained how the tool could be a game-changer for investment analysts.

“Imagine you’re an investment analyst at a Silicon Valley VC firm,” she continued, “You’ve been tasked with evaluating the market potential for civilian supersonic air travel. What do you do?” 

Instead of manually gathering information from multiple sources, which can be time-consuming and overwhelming, she said you could use deep research to get a thorough, well-rounded analysis in a fraction of the time.

The live demo continued with OpenAI’s Neel Ajjarapu showcasing deep research’s effectiveness in compiling reports. He presented a task in which the model analysed the mobile market for language learning. 

“It looked at 29 different sources and gave us a perfectly formatted report, including detailed data on mobile adoption trends,” Neil explained. “You can see every citation the model encountered, along with information about sites it found.” 

In the coming weeks, OpenAI plans to enhance deep research with embedded images, data visualisations, and other analytic outputs. 

Inspired by Google?  

Interestingly, OpenAI’s tool shares not just the name but also bears other similarities with Google Deep Research, which launched last year. Google’s product also acts as a personal AI research assistant, capable of exploring complex topics, synthesising information from multiple sources, and generating comprehensive reports.

However, according to OpenAI, deep research achieved 26.6% on Humanity’s Last Exam—more than double the prior best, o3-mini-high at 13.0%, outperforming DeepSeek R1 and Gemini Thinking.

“OpenAI’s deep research is very good. Unlike Google’s version, which is a summariser of many sources, OpenAI is more like engaging an opinionated (often almost PhD-level!) researcher who follows the lead,” said Ethan Mollick, professor at Wharton. 

“OpenAI deep research is reasoning + tool use agent for deep research. Benchmarks seem really good, but no comparison to Google’s Gemini Deep Research,” said Dylan Patel, founder of Semianalysis. 

Besides Google, Perplexity AI also performs the same task when multiple complex questions are asked. The company has integrated DeepSeek R1 into Perplexity AI as well. It breaks down questions and retrieves the required information from the internet. 

Meanwhile, the DeepSeek chatbot offers both the ability to search and reason. When asked a question, it can conduct a web search to provide relevant answers.