Published: 2025-03-11 16:08:17
Keywords: deep research, ChatGPT, OpenAI, AI agent, multi-step research, internet browsing, data analysis, reasoning
Abstract
OpenAI has launched “deep research” in ChatGPT, an advanced AI capability that conducts comprehensive, multi-step research on the internet to tackle complex tasks in minutes rather than hours. Powered by a version of the upcoming OpenAI o3 model optimized for web browsing and data analysis, this new feature represents a significant advancement toward developing artificial general intelligence (AGI).
What Is Deep Research?
Deep research is the latest “agent” from OpenAI that works independently on your behalf. When you give it a prompt, ChatGPT searches, analyzes, and synthesizes information from hundreds of online sources to create a comprehensive report similar to what a professional research analyst might produce.
Unlike standard ChatGPT interactions, deep research doesn’t just provide quick answers based on its training. Instead, it actively browses the web, reads and interprets text and images, analyzes PDFs, and adjusts its approach based on what it finds. It’s like having a research assistant who can read and understand vast amounts of information quickly and organize it into a coherent report.
How Does It Work?
Deep research was trained using reinforcement learning on challenging browsing and reasoning tasks across various domains. This training taught it to:
- Plan and execute multi-step research strategies
- Backtrack and adjust when needed
- React to new information in real-time
- Browse through user-uploaded files
- Create and refine data visualizations using Python
- Embed both generated graphs and images from websites
- Cite specific sentences or passages from sources
When you select “deep research” in ChatGPT’s message composer and enter your query, the system begins exploring the internet for relevant information. A sidebar appears showing the steps taken and sources used. The research process may take anywhere from 5 to 30 minutes, depending on the complexity of your question.
What Sets Deep Research Apart?
The key difference between deep research and standard GPT-4o is the depth and documentation. While GPT-4o excels at real-time conversations across multiple formats (text, images, etc.), deep research provides extensively researched, well-documented answers with proper citations.
This becomes particularly valuable for:
– Complex business analyses
– Finding obscure information that would require browsing numerous websites
– Academic or medical research
– User experience design
– Detailed shopping research
– Answering difficult knowledge questions
For example, when asked to compile adoption rates of mobile operating systems across developed and developing countries, GPT-4o provides a reasonable summary with limited data. Deep research, however, delivers a comprehensive table with country-specific statistics, detailed analysis, and targeted recommendations backed by multiple sources.
Performance and Capabilities
Deep research shows impressive results on standardized evaluations. On “Humanity’s Last Exam,” which tests AI across expert-level questions in over 100 subjects, the model powering deep research scored 26.6% accuracy—significantly higher than previous models like GPT-4o (3.3%) and OpenAI o1 (9.1%).
It also achieved state-of-the-art performance on GAIA, a benchmark for real-world questions requiring reasoning, multi-modal understanding, web browsing, and tool-use proficiency.
In practical applications, domain experts rated deep research as having automated multiple hours of difficult, manual investigation across fields like chemistry, linguistics, and healthcare.
Current Limitations
Despite its capabilities, deep research still has limitations:
– It may occasionally hallucinate facts or make incorrect inferences
– It can struggle to distinguish authoritative information from rumors
– It sometimes fails to accurately convey uncertainty
– There may be minor formatting errors in reports and citations
– Tasks may take time to begin processing
OpenAI expects these issues to improve rapidly with increased usage and further development.
Availability and Future Plans
Deep research is currently available to ChatGPT Pro users with a limit of 100 queries per month. Plus and Team users will gain access next, followed by Enterprise customers. A faster, more cost-effective version powered by a smaller model is in development and will offer higher rate limits to all paid users.
Currently, deep research can access the open web and uploaded files. Future updates will allow connection to specialized data sources, including subscription-based or internal resources.
Looking ahead, OpenAI envisions combining deep research with its “Operator” capability (which can take real-world actions) to enable ChatGPT to carry out increasingly sophisticated tasks autonomously.
Conclusion
Deep research represents a significant step in AI’s evolution from a conversational tool to an autonomous research assistant capable of independent investigation and analysis. By tackling complex research tasks that would take humans hours to complete, it points toward a future where AI can not only synthesize existing knowledge but potentially help generate new insights as well.