OpenAI Deep Research API
by OpenAI
o3 and o4-mini deep research models over the API
Reviewed by the the30top research team · Last reviewed 2026-10-27
Overview
OpenAI's deep research API exposes o3-deep-research and o4-mini-deep-research, which can use web search, remote MCP servers and file search over your own vector stores to write a report at the level of a research analyst. The final answer carries inline citations. It is built for developers and billed per token, at $2 input and $8 output per million tokens for o4-mini and $10 and $40 for o3, so long runs can get expensive. The o4-mini model does not support function calling or structured outputs.
Our verdict
A capable building block for cited research with internal data. Budget for token costs and plan on engineering time.
Best for
Engineering teams building research agents on OpenAI models that need cited reports from web and internal data.
Avoid if
You want a finished research interface, or a flat monthly price.
Pros
- Can combine web search, MCP servers and your own files
- Inline citations in the final answer
- Cheaper o4-mini option for lighter research
Cons
- Developer product with no analyst interface
- o3 costs $40 per million output tokens
- o4-mini has no function calling or structured outputs
Key capabilities
Frequently asked questions
How much does the OpenAI deep research API cost?
o4-mini-deep-research is $2 per million input tokens and $8 per million output tokens. o3-deep-research is $10 and $40.
Which tools can the deep research models use?
OpenAI lists web search, remote MCP servers and file search over your own vector stores.
Is there a free tier for the deep research API?
No. The o4-mini-deep-research model page lists the Free rate limit tier as not supported.