The O3-Deep-Research model is OpenAI’s flagship agent designed for autonomous, multi-step research workflows. Launched as part of the Deep Research API, it leverages advanced reasoning capabilities to plan, search, analyze, and synthesize information across diverse web sources, delivering expert-level reports within minutes .
Basic Information
The O3-Deep-Research model, unveiled by OpenAI on June 26, 2025, represents a significant leap in automated knowledge work, enabling agentic workflows that autonomously decompose complex tasks, gather web-based data, and synthesize citation-rich outputs . As a flagship offering in the Deep Research API suite—alongside its sibling O4-Mini-Deep-Research—it is designed for professionals in finance, science, policy, and engineering, who demand thorough, accurate, and up-to-date research capabilities .
- Model Name: o3-deep-research-2025-06-26
- Developer: OpenAI
- Release Date: June 26, 2025
- Integration: Fully compatible with ChatGPT Agent framework, supports function calls, webhooks, and browser tools for visual and PDF parsing .
Key Features
- Agentic Task Decomposition: O3-Deep-Research breaks down high-level queries into discrete subtasks, enabling step-by-step data gathering and reasoning.
- Web Grounding: Integrates securely with Bing Search to access authoritative, recent sources, minimizing hallucinations and ensuring data validity .
- Citation-Rich Outputs: Automatically embeds citations in final reports, enhancing transparency and traceability.
- Adaptive Reasoning: Incorporates a private chain of thought framework, allowing dynamic backtracking and hypothesis revision when faced with new information .
Technical Details
The model builds upon the core O3 architecture by integrating specialized modules for deep research tasks:
- Reinforcement Learning with Private Chain of Thought: O3-Deep-Research employs RLHF (Reinforcement Learning from Human Feedback) to internalize multi-step planning, simulating a researcher’s thought process before emitting responses, thereby improving logical coherence and contextual depth.
- Tool Invocation Layer: A dedicated component orchestrates calls to web search, code execution, and data extraction APIs. This layer ensures that each subtask is addressed using the optimal tool, fostering modularity and scalability .
- Safety & Compliance Sandbox: All external web interactions pass through a sandboxed environment that filters out low-credibility domains, applies content safety checks, and logs audit trails for compliance with organizational policies.
Tooling:
- Web Search & Fetch for real-time data retrieval
- Function Calling for custom workflows
- Webhooks to integrate external data sources
- Token Handling: Optimized to manage long-form inputs and expanded outputs (up to 16 K tokens), though very large contexts may incur increased token consumption .
Benchmark Performance
- Humanity’s Last Exam (HLE): Achieved 26.6% accuracy, surpassing DeepSeek’s R1 (9.4%) and GPT-4o (3.3%) .
- GPQA Diamond: Expert-level science benchmark; O3 models have demonstrated 87.7% on advanced questions (o3 full model) .
- Software Engineering (SWE-bench Verified): Scored 71.7%, a significant uplift over non-reasoning counterparts .
- Codeforces: Reached an Elo of 2727, outpacing earlier O1 models (1891).
Model Versions
o3-deep-research-2025-06-26: Debut version with enhanced webhooks and search integration .
Updates:
- Visual browser integration (July 17, 2025)
- Rate-limit adjustments for openAI Official Price :/Pro (+125 full-model +125 lightweight queries per 30 days) and Free/Plus tiers
Limitations
Despite its advances, O3-Deep-Research has notable constraints:
- Latency: The multi-step reasoning and external tool calls introduce higher response times compared to standard LLMs, potentially impacting interactive applications.
- Compute Cost: Reliance on agentic decomposition and chain-of-thought mechanisms increases GPU utilization and operational expenses, requiring careful budget management.
- Hallucinations: While reduced, hallucination rates can still spike in highly ambiguous domains or when encountering paywalled or low-quality sources, necessitating human oversight.
- Scope Boundaries: The model may struggle with non-public or proprietary data without specialized connectors, limiting its applicability in closed-access environments.
How to call O3-Deep-Research API from CometAPI
O3-Deep-Research API Pricing in CometAPI，20% off the official price:
| Input Tokens | $8.00 |
|---|---|
| Output Tokens | $32.00 |
Required Steps
- Log in to cometapi.com. If you are not our user yet, please register first
- Get the access credential API key of the interface. Click “Add Token” at the API token in the personal center, get the token key: sk-xxxxx and submit.
- Get the url of this site: https://api.cometapi.com/
Use Method
- Select the “
o3-deep-research”/ “o3-deep-research-2025-06-26” endpoint to send the API request and set the request body. The request method and request body are obtained from our website API doc. Our website also provides Apifox test for your convenience. - Replace <YOUR_API_KEY> with your actual CometAPI key from your account.
- Insert your question or request into the content field—this is what the model will respond to.
- . Process the API response to get the generated answer.
CometAPI provides a fully compatible REST API—for seamless migration. Key details to Response API doc:
- Base URL: https://api.cometapi.com/v1/responses
- Model Names: “
o3-deep-research”/ “o3-deep-research-2025-06-26“ - Authentication:
Bearer YOUR_CometAPI_API_KEYheader - Content-Type:
application/json.
API Integration & Examples
Developers can integrate O3-deep-research via the Responses API. A minimal example:
curl
--location 'https://api.cometapi.com/v1/responses' \
--header 'Authorization: Bearer sk-xxxxx' \
--header 'Content-Type: application/json' \
--data '{
"model": "o3-deep-research-2025-06-26",
"stream": true, "reasoning": { "summary": "detailed" },
"tools": [ { "type": "web_search_preview" } ],
"input": "who are you"
}'See also
