OpenAI announced a new feature called "deep research" on 2 February 2025, positioning it as a tool that performs extended web searches and synthesises findings into comprehensive reports [source]. The feature is available to ChatGPT Pro subscribers and operates through the o1 reasoning model.

According to the announcement, deep research can spend "minutes or more" gathering information across multiple web sources before compiling results. OpenAI describes the process as the model creating a research plan, executing searches, and iteratively refining its approach based on what it finds. Users receive a final report with citations.

The feature represents a shift from single-query responses to autonomous multi-step research sessions. OpenAI states that deep research can "visit dozens of sites" and "follow up on what it learns" without further user input during the research phase.

No independent verification of output accuracy or citation reliability was provided in the announcement. The feature's behaviour when encountering conflicting sources, paywalled content, or outdated information remains undocumented. OpenAI did not specify how the model determines when research is complete or how it prioritises sources.

The announcement follows a pattern of providers expanding model autonomy in information-gathering tasks. Similar features have been introduced by other providers, though comparative performance data is not available.

ChatGPT Pro costs $200 per month. OpenAI indicated the feature will expand to other subscription tiers but provided no timeline. The company published example research reports on topics including graduate programme selection and CRISPR technology applications.

Users seeking to verify deep research outputs will need to manually review cited sources, as the announcement contains no discussion of hallucination mitigation or citation validation mechanisms.