OpenAI o3-miniTranslation site

OpenAI What is o3-mini

The OpenAI o3-mini was developed by theA small inference model from OpenAI Inc... The model is designed to meet the needs of developers and users for efficient, low-cost AI solutions.

o3-mini inherits the strengths of OpenAI's previous models and makes significant improvements in inference capability, feature richness, cost-effectiveness, and security. It supports function calls, streaming, structured output, etc., and can be combined with a search function to provide up-to-date answers and links to relevant web resources. In addition, o3-mini is optimized for problems in STEM fields (science, technology, engineering, and mathematics) and performs well.

As OpenAI continues to promote the innovation and development of AI technology, the launch of o3-mini will further expand the application scenarios of AI technology and bring users a more high-quality and efficient AI experience.

OpenAI o3-mini model features

1. reasoning ability::

   • The o3-mini has a powerful reasoning ability that allows it to retrieve relevant information from the web and combine it with existing knowledge to make more precise judgments and answers. This analytical ability allows it to mimic the human thought process and find solutions more effectively.
   • 在不同推理强度下，o3-mini能够灵活适应不同的任务需求。在高推理强度下，它的表现优于o1-mini和o1；在中等推理强度下，与o1相当，但在数学、program和科学领域以更快的响应速度实现了与o1相当的性能；在低推理强度下，其表现与o1-mini相当。

2. cost-effectiveness::

   • Compared to OpenAI's previous model, o3-mini has a significant price reduction. Compared to o1-mini, its price is reduced by 631 TP4T. pricing is $1.10 per million input tokens and $4.40 per million output tokens. Although the o3-mini is still more expensive compared to the DeepSeek R1, it has made a significant improvement in price/performance ratio.

3. feature-rich::

   • o3-mini supports function calls, streaming, structured outputs, etc. Developers can choose the reasoning strength according to their needs, balancing the depth of thinking and response speed.
   • It now supports integration with the search function, which can provide up-to-date answers and links to relevant web resources. This signals that OpenAI is gradually integrating search functionality into itsinference modelCenter.

4. safety::

   • OpenAI used a thoughtful alignment approach when training o3-mini to ensure that the content it generates is safer, more ethical, and reduces the risk of the model generating bad or harmful responses.

OpenAI o3-mini Application Scenarios

1. Developer Tools::

   • o3-mini is OpenAI's first small inference model to support developer-requested features, allowing developers to leverage its powerful inference capabilities and rich functionality to optimize the performance and efficiency of their applications.

2. STEM fields::

   • o3-mini focuses on problems related to STEM fields such as programming, math, and science, as well as logical reasoning problems. It excels when it comes to more technical and complex tasks and helps developers solve challenges in code writing, mathematical computation, engineering design, and more.

3. Education and training::

   • o3-mini can show users their thinking process by clicking the "Reasoning" button. This feature will greatly enhance the users' interest in learning and help them to acquire more in-depth knowledge in a relaxing and enjoyable atmosphere.

OpenAI o3-mini User Access and Restrictions

1. access authority::

   • ChatGPT Plus, Team and Pro users are the first to use o3-mini, with enterprise users accessing it a week later. For API users, o3-mini has been deployed to the Chat Completions API, Assistants API and Batch API for developers in tiers 3-5.

2. speed limit::

   • The rate limit for ChatGPT Plus and Team users has been increased from 50 messages per day (with o1-mini) to 150 messages per day (with o3-mini). o3-mini is available to ChatGPT Pro users with unlimited access. free users can also experience o3-mini through the "Reasoning" option. Free users can also experience o3-mini through the "Reasoning" option.