OpenAI has announced the launch of Deep Research, a groundbreaking AI agent designed to perform multi-step research on the internet. This new feature, available in ChatGPT, allows users to offload complex online research tasks, synthesizing vast amounts of information in a matter of minutes.
Built on an advanced version of the upcoming OpenAI o3 model, Deep Research excels in data analysis, knowledge synthesis, and web-based reasoning. The model independently searches, analyzes, and compiles data, producing research analyst-level reports with clear citations and references.
The feature is currently available to ChatGPT Pro users, with Plus and Team users set to gain access next.
Why OpenAI Developed Deep Research
Deep Research was built for professionals and researchers who require in-depth knowledge across fields such as finance, science, policy, and engineering. The tool is also designed for consumers seeking hyper-personalized recommendations, making it a valuable asset for thorough product comparisons on purchases like cars, appliances, and technology.
Unlike traditional AI models that provide brief summaries, Deep Research is capable of:
- Browsing and analyzing hundreds of online sources
- Generating well-documented reports
- Providing structured citations for fact verification
- Synthesizing non-intuitive and hard-to-find insights
This advancement represents a major step toward OpenAI’s long-term goal of Artificial General Intelligence (AGI), which includes the ability to generate new knowledge rather than simply retrieving existing data.
How Deep Research Works
Users can activate Deep Research in ChatGPT by selecting the “Deep Research” option in the message composer. After entering a query, such as a competitive analysis of streaming services or a report on electric vehicle adoption trends, the AI begins an in-depth investigation.
Key Features:
- Multi-Step Research: Deep Research autonomously plans and executes complex research trajectories, adjusting based on real-time findings.
- File and Spreadsheet Support: Users can upload files or spreadsheets to enhance contextual understanding.
- Live Tracking: A sidebar provides updates on research steps, sources used, and analysis progress.
- Comprehensive Reports: Within 5 to 30 minutes, ChatGPT delivers a detailed research report, complete with citations.
- Upcoming Enhancements: Future updates will introduce embedded images, data visualizations, and analytical graphs to enhance reports.
Unlike GPT-4o, which excels at real-time multimodal interactions, Deep Research is designed for extensive, domain-specific inquiries requiring meticulous fact-checking and synthesis.
Benchmark Performance: A New Standard in AI Research
Deep Research significantly outperforms previous AI models in expert-level research evaluations, demonstrating superior reasoning and accuracy.
Humanity’s Last Exam Performance
In the Humanity’s Last Exam, a rigorous AI benchmark testing expert-level reasoning across 100+ subjects, Deep Research achieved a record 26.6% accuracy, outperforming all major AI models.
| Model | Accuracy (%) |
|---|---|
| GPT-4o | 3.3 |
| Grok-2 | 3.8 |
| Claude 3.5 Sonnet | 4.3 |
| Gemini Thinking | 6.2 |
| OpenAI o1 | 9.1 |
| DeepSeek-R1 | 9.4 |
| OpenAI o3-mini (high) | 13.0 |
| OpenAI Deep Research | 26.6 |
Deep Research excels in chemistry, humanities, social sciences, and mathematics, using its advanced reasoning capabilities to search and interpret complex data more effectively than prior models.
GAIA Benchmark Results
Deep Research has also set a new state-of-the-art (SOTA) record on GAIA, a public benchmark that tests AI’s ability to handle real-world, multi-step research tasks.
| Test Level | Previous SOTA (%) | Deep Research Pass@1 (%) | Deep Research Consensus@64 (%) |
|---|---|---|---|
| Level 1 | 67.92 | 74.29 | 78.66 |
| Level 2 | 67.44 | 69.06 | 73.21 |
| Level 3 | 42.31 | 47.6 | 58.03 |
| Average | 63.64 | 67.36 | 72.57 |
These results confirm that Deep Research is not only faster and more efficient but also capable of tackling highly technical research challenges with greater precision.
Challenges and Future Improvements
Despite its advanced capabilities, Deep Research is still in its early stages and has some limitations:
- Fact Hallucination: While significantly reduced, occasional misinterpretations or inaccurate inferences may occur.
- Confidence Calibration Issues: The model may not always express uncertainty accurately, potentially leading to overconfidence in some responses.
- Formatting & Citation Errors: Minor formatting inconsistencies and citation placement issues are being refined.
As Deep Research evolves, OpenAI plans to enhance reliability, improve accuracy, and expand access to additional specialized data sources.
Availability and Access
Who Can Use Deep Research?
- Pro Users: Available now, with a limit of 100 queries per month.
- Plus & Team Users: Access rolling out soon.
- Enterprise Users: Future release planned.
Deep Research is currently unavailable in the UK, Switzerland, and EEA, but OpenAI is actively working on expanding regional access.
Upcoming Enhancements
- A More Cost-Effective Version: A smaller, faster, and more efficient Deep Research model will soon become available to all paid users.
- Mobile & Desktop Integration: Deep Research will roll out to ChatGPT’s mobile and desktop apps within the next month.
- Expanded Data Access: Future updates will integrate subscription-based and internal data sources, making research even more comprehensive.
The Future of AI-Powered Research
OpenAI’s Deep Research represents a major leap forward in AI’s ability to conduct independent, multi-step reasoning. By combining intelligent data synthesis, real-time web browsing, and advanced reporting, OpenAI is paving the way for more autonomous AI-powered research tools.
Looking ahead, OpenAI envisions even more sophisticated AI agents capable of performing asynchronous online research and real-world task execution. The integration of Deep Research with OpenAI’s upcoming agent, Operator, promises to redefine AI’s role in research, automation, and decision-making.
For now, Deep Research is set to revolutionize how professionals, researchers, and consumers gather and synthesize information transforming hours of work into minutes.