o3-deep-research and o4-mini-deep-research

2025-06-26 · Publisher : OpenAI · Domain : Artificial Intelligence

OpenAI opens two deep research models in the API, o3-deep-research and o4-mini-deep-research, derived from the o3 and o4-mini reasoning models to analyze and synthesize web and private sources.

What changes

  • API versions of the deep research engines in ChatGPT, for multi-step research on the Internet.
  • Private sources can be connected through MCP connectors.
  • o4-mini-deep-research is the faster and cheaper variant.
  • They work in the Responses API, with web search as a tool and webhook notifications (launched the same day).
  • Positioned as specialized models: they replace neither o3 nor o4-mini for general-purpose reasoning, but target long, well-documented reports.
  • Designed for long runs that are handled in the background rather than as an interactive request.

### Specifications - Context 200,000 tokens, maximum output 100,000 tokens, text and image input, text output. - Knowledge cutoff: June 1, 2024. - API identifiers: o3-deep-research and o4-mini-deep-research (snapshots 2025-06-26). ### Pricing Pricing per million tokens (USD), according to the documentation. | Model | Input | Cached input | Output | | --- | --- | --- | --- | | o3-deep-research | 10 | 2.50 | 40 | | o4-mini-deep-research | 2 | 0.50 | 8 | ### Benchmarks No scores published by the vendor for these models in the sources consulted. The results of the ChatGPT version (26.6% on Humanity's Last Exam) appear in the Deep research entry of February 2, 2025. ### Key takeaways These models make it possible to integrate agentic document research into enterprise applications, with internal sources via MCP. Requests are long and costly, and should be handled as background tasks.

Links

Tags : LLM, OpenAI, o-series, Deep research, Agent