ChatGPT deep research
Ask a question and it spends minutes reading the web, then returns a long report with citations.
By OpenAI · launched 2025-02-03 · last checked 2026-11-01
Why 4: it browses for five to thirty minutes, cross-checks sources and writes a cited report on its own for you to review; you can steer it, but it acts only when asked.
What it does. Deep research browses for roughly five to thirty minutes, reading and cross-checking sources, and writes a cited report. A February 2026 update moved it to a GPT-5.2-based model, let users restrict it to chosen sites, connect apps and data through MCP servers, and steer it mid-run. When OpenAI published its BrowseComp benchmark in April 2025, deep research solved 51.5% of its hard-to-find questions; the paper noted it had been trained for that kind of task.
Where it falls short. OpenAI acknowledges it sometimes makes factual errors and wrong inferences. An April 2026 study found that 3–13% of citation URLs from commercial models and deep research agents had no record in the Wayback Machine and likely never existed, and that deep research agents fabricated URLs at higher rates.
What it can do
Incidents
- None logged.
Closest fiction
Her — Samantha reading at superhuman speed. She also knew which books were real.
Related research agents
- Gemini Deep Research Google · autonomy 4/5
- Perplexity Deep Research Perplexity · autonomy 4/5
History
- 2026-11-01 Added to the index.
Sources
- ChatGPT Deep Research Wikipedia
- BrowseComp: A Simple Yet Challenging Benchmark for Browsing Agents arXiv (OpenAI), 2025-04-16
- Detecting and Correcting Reference Hallucinations in Commercial LLMs and Deep Research Agents arXiv (Rao, Wong and Callison-Burch), 2026-04-03
← All agents in the index · Autonomy is rated on what the product does today, on a 1–5 scale. Every fact links to a source.