OpenAI Unveils ChatGPT Deep Research Agent for In-Depth AI Analysis

OpenAI has unveiled ChatGPT agent Deep Research, a new AI feature designed to assist users with in-depth, complex multi-source research. Targeted at professionals in finance, science, policy, and engineering, as well as consumers researching major purchases like cars, appliances or furniture, Deep Research offers a more thorough and analytical approach compared to traditional chatbot responses.

How Deep Research Works

• Available now for ChatGPT Pro users (limited to 100 queries per month). Plus and Team users will gain access soon, followed by Enterprise. OpenAI plans to roll out Deep Research to Plus users within the next month and expects to significantly increase query limits for paid users in the near future.

• Web-only at launch, with mobile and desktop app integration coming later this month.

• To use Deep Research, users must select it in the ChatGPT composer before entering a query. The feature also supports file and spreadsheet uploads for more detailed analysis.

• Takes 5 to 30 minutes per query, notifying users once results are ready.

• Supports file and spreadsheet uploads for more tailored analysis.

• Currently outputs are text-only, but OpenAI plans to add data visualizations, images, and analytic outputs in future updates.

• Future integrations may allow access to more specialized data sources, including subscription-based and internal resources.

• OpenAI says they are actively working to expand access to users in the United Kingdom, Switzerland, and the European Economic Area.

Powered by OpenAI’s Advanced Reasoning Model

Deep Research leverages a specialized version of OpenAI’s o3 reasoning AI, optimized for web browsing and data analysis. This model was trained using reinforcement learning, a technique where AI improves through trial and error, receiving virtual 'rewards' for getting closer to a goal, on real-world tasks requiring Python tools and internet searches.

• Searches, interprets, and analyzes large datasets from websites, PDFs, and user-uploaded files.

• Generates and embeds graphs, images, and cited references in responses.

• OpenAI tested ChatGPT Deep Research using Humanity’s Last Exam, a benchmark designed to assess AI reasoning across more than 3,000 expert-level questions spanning fields like law, medicine, engineering, and mathematics. The o3 model powering Deep Research achieved an accuracy of 26.6%, significantly outperforming competitors—Gemini Thinking (6.2%), Grok-2 (3.8%), and GPT-4o (3.3%). OpenAI emphasizes that this benchmark is intentionally difficult, aiming to stay ahead of rapid AI advancements.

For more benchmarks and details, please visit OpenAI’s announcement.

Accuracy Challenges & Future Improvements

While OpenAI emphasizes citation transparency and clear documentation, Deep Research is not immune to errors. The model:

• Struggles to differentiate authoritative sources from rumors.

• Fails to indicate uncertainty in some cases.

• May produce incorrect citations or formatting errors.

Given AI’s history of hallucinations and misinformation, OpenAI’s approach is aimed at providing verifiable research, but users are encouraged to fact-check results rather than relying on them blindly.

What This Means

With Deep Research, OpenAI is pushing ChatGPT beyond quick summaries into more serious research and analysis. If successful, this could disrupt traditional search engines by offering structured, multi-source research instead of simple search results. Professionals in fields like finance, science, and policy may find it a valuable tool for gathering and analyzing information more efficiently for deeper insights across multiple domains. However, its potential impact depends on accuracy—while OpenAI has built in citations and documentation, Deep Research is still prone to errors, misinterpretations, and formatting inconsistencies.

At the same time, this shift raises questions about the cost and accessibility of AI-powered research. If high-quality research assistance becomes dependent on paid AI services, it could influence how knowledge is accessed and trusted.

Interestingly, Google announced an AI tool with the exact same name just two months ago, signaling a growing competition in AI-powered deep research solutions. Whether OpenAI’s implementation proves superior remains to be seen, but the race to create the most effective, AI-driven research assistant is clearly underway.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing, image, and idea-generation support from ChatGPT, an AI assistant. However, the final perspective and editorial choices are solely Alicia Shapiro’s. Special thanks to ChatGPT for assistance with research and editorial support in crafting this article.