OpenAI Releases O3 Deep Research Models
OpenAI released o3-deep-research and o4-mini-deep-research on June 24, 2025. These models are optimized for deep analysis tasks via the v1/responses API.
Official Release Details
OpenAI officially released two new model variants on June 24, 2025: o3-deep-research and o4-mini-deep-research. The source identifies these as deep research variants of the existing o-series reasoning models. According to the official changelog, these models are optimized for deep analysis and research tasks. Specific dated versions, o3-deep-research-2025-06-26 and o4-mini-deep-research-2025-06-26, were also listed in the release notes.
- Models released: o3-deep-research and o4-mini-deep-research.
- API endpoint: v1/responses.
- Input modalities: text and image.
- Output modality: text.
Associated Feature Updates
The release announcement coincided with several platform-level updates. OpenAI added support for the web search tool, noting that pricing for this tool was reduced and simplified. Additionally, the platform now supports async event handling using webhooks. The source does not specify how the web search tool interacts with the new deep research models or provide technical details on the webhook implementation.
- Added support for the web search tool.
- Reduced and simplified pricing for the web search tool.
- Added support for async event handling with webhooks.
Evaluation Procedure and Limitations
Engineers can evaluate these models by testing them against specific research tasks to measure output quality. A practical procedure involves defining a set of complex analytical prompts, sending requests via the v1/responses API, and comparing the resulting text outputs against established ground truth data. It is important to note that open weights are not confirmed for these models. The source does not provide context window limits, maximum output tokens, or specific performance benchmarks. Users should validate results independently rather than assuming comparative improvements over previous iterations.
- Test: Submit diverse research prompts via the v1/responses API.
- Measure: Assess the accuracy and depth of the generated text responses.
- Compare: Validate findings against known correct answers or expert reviews.
Primary sources
Frequently asked questions
What is the primary purpose of the o3-deep-research model?
The o3-deep-research model is a variant of the o-series reasoning models optimized specifically for deep analysis and research tasks. It accepts text and image inputs and produces text output.
Are the model weights available for local installation?
The source indicates that open weights are not confirmed. Therefore, it cannot be stated that local installation is possible or impossible based solely on this release information.