OpenAI Launches o3-mini: A New Milestone in High-Performance AI

At the end of January 2025, OpenAI introduced the latest generation of its small AI model—o3-mini. This model not only inherits the strengths of its predecessors but also brings significant improvements in performance and functionality. This article explores the key features, applications, and comparisons of o3-mini with other models.

Core Features of o3-mini

Outstanding Performance in STEM Fields

o3-mini excels in science, technology, engineering, and mathematics (STEM):

• Achieves 83.6% accuracy in high-intensity reasoning mode on AIME 2024 competition math problems
• Scores 77% accuracy in doctoral-level science question evaluations
• Solves over 32% of research-level math problems when using Python tools

Faster Response Speed

Compared to previous models, o3-mini delivers superior performance:

• Response time reduced by 24% (from 10.16s to 7.7s)
• First output delay reduced by 2500ms
• Supports streaming output for a smoother experience

Developer-Friendly Features

o3-mini offers multiple practical features for developers:

• Function Calling support
• Structured Outputs for better data handling
• Developer message support
• Three reasoning intensity options (Low, Medium, High) for adjustable performance

User Access and Permissions

Paid User Access

• Available immediately for ChatGPT Plus, Team, and Pro users
• Enterprise users will gain access in February 2025
• Daily message limit increased to 150 for Plus and Team users (previously 50)

Free User Access

• First-time availability of reasoning models for free users
• Can be accessed via the message editor by selecting the “Reason” option
• Supports response regeneration

Security and Performance Evaluation

Enhanced Security Features

• Uses deliberative alignment for improved safety
• Outperforms GPT-4o in security tests
• Underwent rigorous external red team testing and security assessments

Performance Evaluation Results

• Expert testers preferred o3-mini responses 56% of the time
• Major errors in complex real-world tasks reduced by 39%
• Showcased strong performance in programming competition evaluations

Conclusion

The release of o3-mini marks a new balance between AI performance and practicality. Whether for general users or developers, this model offers significant benefits, especially in technical fields requiring accuracy and fast responses.

Frequently Asked Questions

Q: What are the main differences between o3-mini and the o1 model? A: o3-mini performs similarly to o1 in STEM fields but offers faster response times and more developer-friendly features.

Q: Does o3-mini support vision-related tasks? A: No, vision-related tasks still require the OpenAI o1 model.

Q: How should I choose the appropriate reasoning intensity? A:

• Use Low for simple tasks to maximize speed.
• Use High for complex tasks to improve accuracy.