OpenAI has launched Deep Research in ChatGPT, a powerful new capability designed to conduct multi-step research on the internet for complex queries. This new agent automates what would typically take a human hours, delivering comprehensive, well-documented reports in just minutes.
What is Deep Research?
Deep Research is an advanced research assistant optimized for web browsing and data analysis, powered by OpenAI’s upcoming o3 model. It can independently search, analyze, and synthesize information from hundreds of sources, including text, images, and PDFs, providing users with structured, high-quality insights.
This tool is particularly useful for professionals in finance, science, policy, and engineering, as well as for consumers making well-researched purchasing decisions.
How It Works
Users simply select Deep Research in ChatGPT, input their query, and let the tool do the work. It carefully tracks its research steps, cites sources, and summarizes findings.
The process can take anywhere from 5 to 30 minutes, allowing users to focus on other tasks while waiting for a notification once the research is complete. Soon, reports will also include embedded images and data visualizations.
Why It Matters
Unlike GPT-4o, which excels in real-time conversations, Deep Research is tailored for in-depth analysis, making it ideal for tasks requiring extensive information gathering and verification. It builds on OpenAI’s reinforcement learning techniques and has been rigorously tested on real-world challenges.
Performance Benchmarks
Deep Research has set a new benchmark in AI-driven research:
- Humanity’s Last Exam: It achieved a record 26.6% accuracy on expert-level questions across various subjects, far surpassing other AI models.
- GAIA Benchmark: It reached a new state-of-the-art performance, excelling in real-world multi-modal reasoning and browsing tasks.
| 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 (medium)* | 10.5 |
| OpenAI o3-mini (high)* | 13.0 |
| OpenAI deep research** | 26.6 |
* Model is not multi-modal, evaluated on text-only subset. **with browsing + python tools
Limitations & Availability
While significantly reducing hallucinations and errors compared to earlier models, Deep Research still struggles with distinguishing authoritative sources from unreliable ones.
At launch, it’s available to ChatGPT Pro users with a limit of 100 queries per month, with Plus and Team users gaining access next. OpenAI is working on expanding availability and improving speed and cost-efficiency.
The Future of AI Research
Deep Research will soon integrate with more specialized data sources, making it even more powerful. OpenAI also envisions combining it with Operator, a future tool designed to take real-world actions based on research findings, further automating complex workflows.
For now, Deep Research is available on ChatGPT Web, with mobile and desktop support rolling out soon.
