ChatGPT Gains Advanced Agentic Capability for In-Depth Research

OpenAI has introduced a groundbreaking feature called Deep Research, empowering ChatGPT to handle complex, multi-step research tasks with remarkable efficiency. This new capability enables AI to complete tasks in minutes, typically taking human researchers hours or days.

A Milestone Toward Artificial General Intelligence (AGI)

OpenAI regards Deep Research as a significant step toward Artificial General Intelligence (AGI). According to the company, synthesising knowledge is a crucial prerequisite for creating new knowledge. Deep Research exemplifies this principle by autonomously gathering, analyzing, and structuring information into insightful reports.

AI-Powered Research: Enhancing Analytical Capabilities

Deep Research leverages an advanced variant of OpenAI’s upcoming “o3” model to streamline and enhance research workflows. Users can simply input a query, and the AI autonomously browses, verifies, and compiles information from hundreds of online sources. The results mirror the quality of a professional research analyst’s report.

Key Features of Deep Research:

• Comprehensive Reports: Produces detailed insights across industries such as finance, science, policymaking, and engineering.
• Cited Sources: Ensures transparency by including full citations, allowing users to verify information easily.
• Automated Competitive Analysis: Conducts deep comparisons, making it ideal for business intelligence and policy evaluations.
• User-Friendly Interface: Accessible directly within ChatGPT, with an intuitive research workflow.
• This functionality can benefit many users, from companies conducting market research to individuals seeking personalized product recommendations.

How Deep Research Works

Users can access Deep Research via the ChatGPT interface by selecting the feature in the message composer. The AI then embarks on a multi-step research process that can take 5 to 30 minutes, depending on the complexity of the request.

Research Process Overview:

• Data Collection: The AI autonomously scans and retrieves relevant information from online sources.
• Analysis & Synthesis: It evaluates and organizes the data, extracting key insights.
• Comprehensive Reporting: Generates a well-documented, easy-to-read report with citations.
• User Updates: A sidebar provides real-time updates on progress and consulted sources.

OpenAI is actively enhancing Deep Research to include data visualizations, graphs, and embedded media, making insights more digestible and engaging.

Deep Research vs. GPT-4o: A Shift Toward Depth and Accuracy

Unlike GPT-4o, which focuses on real-time, multimodal interactions, Deep Research prioritizes accuracy, detailed citations, and extensive analysis. This shift caters to users who require well-documented, research-grade content rather than fast, summarized responses.

Advanced AI for Real-World Problem Solving

Deep Research is trained using reinforcement learning, allowing it to navigate real-world browsing scenarios and complex reasoning tasks. Its capabilities extend beyond standard AI responses by integrating tools like:

• Python for Data Analysis: Generates and iterates on visualizations.
• Document Processing: Analyzes user-uploaded files.
• Media Embedding: Includes images and referenced web pages in responses.

These advanced functionalities make Deep Research a powerful assistant for scientists, policymakers, engineers, and business leaders tackling intricate challenges.

Benchmark Performance: Setting New Records

OpenAI evaluated Deep Research against Humanity’s Last Exam, a rigorous test comprising over 3,000 expert-level questions spanning diverse fields like rocket science, linguistics, and ecology. The results were groundbreaking:

• Model
• Exam Accuracy (%)
• GPT-4o
• 3.3%
• Grok-2
• 3.8%
• Claude 3.5 Sonnet
• 4.3%
• OpenAI o1
• 9.1%
• DeepSeek-R1
• 9.4%
• Deep Research
• 26.6% (with browsing + Python tools)

Additionally, Deep Research achieved state-of-the-art performance on the GAIA benchmark, securing a 72.57% score. This evaluation measures AI competence in reasoning, multimodal fluency, and tool-use proficiency.

Challenges and Limitations

While Deep Research represents a major technological leap, OpenAI acknowledges certain limitations:

• Occasional Hallucinations: The AI sometimes generates incorrect inferences, though at a reduced rate compared to previous models.
• Source Differentiation Issues: Struggles to distinguish between authoritative sources and speculative content.
• Confidence Calibration Problems: Can display undue certainty about uncertain findings.
• Minor Formatting Errors: Some reports may contain citation inconsistencies or slight delays in generation.

Despite these hurdles, OpenAI is actively refining Deep Research through iterative improvements and user feedback.

Gradual Rollout & Future Developments

OpenAI is gradually rolling out Deep Research to Pro users with a limit of 100 queries per month. Expansion plans include Plus, Team, and Enterprise tiers, followed by ChatGPT’s mobile and desktop platforms availability.

Upcoming Enhancements:

• Integration with Subscription-Based Data Sources: To provide richer, more personalized insights.
• Cross-Platform Access: Extending usability across various devices.
• Collaboration with OpenAI’s Operator: A chatbot that can execute real-world actions with online Research.

The Future of AI-Driven Research

Deep Research marks a transformative step in AI-assisted information gathering. OpenAI redefines how professionals and everyday users access and interpret information by automating complex research tasks. This powerful tool is poised to revolutionize data-driven decision-making across industries as improvements continue.

With its advanced capabilities, Deep Research isn’t just about retrieving information—it’s about unlocking new knowledge.