← Models

o3-deep-research

OpenAI · o-series

OpenAI's highest-capability deep research model for long, source-heavy investigations over web and private data.

Type
language
Context
200K tokens
Max Output
100K tokens
Status
current
Input
$10/1M tok
Output
$40/1M tok
API Access
Yes
License
proprietary
deep-research reasoning web-search mcp file-search analysis
Released June 2025 · Updated April 8, 2026

Overview

Freshness note: Model capabilities, limits, and pricing can change quickly. This profile is a point-in-time snapshot last verified on April 8, 2026.

o3-deep-research is OpenAI’s higher-end deep research model. OpenAI describes it as the most powerful model in this category and positions it for complex, multi-step research tasks that pull together web sources, private data, and longer-form analytical synthesis.

This is not simply “o3 with web search attached.” The current deep research guide frames these models as a distinct workflow that can browse, search, fetch, analyze, and synthesize many sources into a report closer to research-analyst output than to ordinary chat.

Capabilities

o3-deep-research is optimized for source-heavy investigation. OpenAI’s current guide says deep research models can search the public web, use remote MCP servers, search internal vector stores, and optionally use code interpreter for analysis. That makes the model suitable for legal research, market analysis, technical landscape briefs, regulatory tracking, and internal reporting over large document sets.

The practical value is not just depth of reasoning. It is the ability to work across retrieval plus analysis in a more agentic loop than a normal single-response model.

Technical Details

Current published limits:

  • Context window: 200,000 tokens
  • Max output: 100,000 tokens

OpenAI’s current model docs support text input/output and image input, but not audio or video. Streaming is supported. Function calling and structured outputs are not supported on this model, which matters if your downstream system expects rigid schemas.

OpenAI’s deep research guide also notes a key operational constraint: these models require at least one data source such as web search, remote MCP servers, or file search over vector stores. They are designed to research, not to sit idle inside a plain prompt-only workflow.

Pricing & Access

Published pricing (per 1M tokens):

  • Input: $10.00
  • Output: $40.00

Access is through OpenAI API deep research workflows. OpenAI recommends background mode for long-running tasks, which is sensible because these requests can run for minutes and involve multiple tool calls.

Best Use Cases

Use o3-deep-research for high-value investigations where source coverage, synthesis quality, and analytical depth matter more than low cost. It is a strong fit for regulated-domain research, executive brief preparation, competitor and market analysis, and internal knowledge synthesis over many documents.

Comparisons

  • o4-mini-deep-research (OpenAI): Cheaper and faster deep research route; o3 is the higher-capability option.
  • o3 (OpenAI): Better for general reasoning without the full deep research workflow and tool assumptions.
  • GPT-5.4 (OpenAI): Strong general professional model, but not as explicitly specialized for source-heavy deep research loops.