Deep research

2025-02-02 · Publisher : OpenAI · Domain : Artificial Intelligence

OpenAI launches deep research in ChatGPT, an agent that carries out multi-step web research with a version of o3 optimized for browsing and data analysis.

What changes

  • Autonomous agent that explores, analyzes and synthesizes hundreds of web sources (text, images, PDFs) to produce an analyst-level report.
  • Trained by reinforcement learning on real tasks requiring browsing and Python.
  • Each report is documented with citations and a summary of the reasoning.
  • A lightweight version based on o4-mini is added on April 24, 2025 to increase quotas.
  • The o3-deep-research and o4-mini-deep-research variants arrive in the API on June 26, 2025 (dedicated entry).

### Specifications - Model: a version of o3 dedicated to web browsing and analysis; API identifier not applicable at announcement. - Availability at launch: ChatGPT Pro users (100 queries per month), then Plus and Team, then Enterprise; Plus on February 25, 2025. - Typical duration: several tens of minutes per query. - Context window, maximum output and knowledge cutoff: not disclosed. ### Pricing Pricing not disclosed at launch (included in the ChatGPT Pro subscription, monthly quotas). ### Benchmarks Scores published by the vendor. | Benchmark | Result | | --- | --- | | Humanity's Last Exam | 26.6% (new record claimed) | | GAIA | state of the art claimed (detailed values in chart) | ### Key takeaways Deep research automates monitoring and long-form documentary research, with verifiable sources. The vendor points out limits (possible factual errors, difficulty distinguishing reliable sources).

Links

Tags : LLM, OpenAI, Agent, Deep research, o3