[OpenAI] What is o4-mini? Explaining the main features, usage, and fee structure

A thorough explanation of the latest small AI model, the o4-mini: High-speed, low-priced, high-performance!

OpenAI has released a new small AI model, the o4-mini. Unlike the flagship model "O3", its appeal is its high speed and low price. In this article, we will explain in detail the features of the O4-mini, its differences from the O3, its usage situation, its safety, and how to use it.

table of contents

1. Overview of o4-mini

2. Key Features

3. About o4-mini-high

4. How to use and fee structure

5. API usage and fees

6. Use with Azure and GitHub Copilot

7. Benchmark performance

8. The benefits of choosing o4-mini

9. Differences from o3

10. Safety initiatives

11. summary

12. Overview of o4-mini

The o4-mini is a high-speed, low-cost, compact AI model developed by OpenAI. Unlike the "o3", it is designed to be easy to use.

2. Key Features

• Fast inference processing
• Excellent cost performance
• Small size, yet highly efficient (especially mathematical processing, code generation, image processing)
• As a successor to the o3-mini, it has improved balance of speed, price and performance.

3. About o4-mini-high

o4-mini-high is a faster model of o4-mini and is ideal for image inference and coding.

4. How to use and fee structure

You can also try out some of the features of the o4-mini with the free plan.

• Free plan: Select the "Think" option on the ChatGPT text writing screen. Features are limited.
• Paid Plans (Plus/Pro/Team): Select o4-mini from the ChatGPT model selection menu. Automatic switching from o3-mini is available.
• Enterprise/For educational institutions (Edu): Available soon.

5. API usage and fees

Developers can use the Chat Completions API and the Responses API (pre-confirmation is required for organizational use). The Responses API will also provide a function to improve performance by maintaining thought processes in the future.

item	Fee (per million tokens)
input	$1.10 (approx. 165 yen)
Cache input	$0.275 (approx. 41 yen)
output	$4.40 (approx. 660 yen)

Rate limiting is the same as in the previous model. For more information, please refer to the official OpenAI documentation.

6. Use with Azure and GitHub Copilot

It is also available in Azure OpenAI Service and GitHub Copilot on Microsoft Azure.

7. Benchmark performance

The o4-mini has received high praise in a variety of benchmarks.

• Mathematical Ability: High accuracy (over 90%) with AIME benchmark.
• Coding ability: High accuracy with SWE-Bench, Aider Polyglot, etc.
• Visual and multimodal capabilities: High accuracy with MMMU, MathVista, ChartXiv, etc.
• Instruction tracking ability: Practical accuracy with Scale MultiChallenge.

8. The benefits of choosing o4-mini

• Large-scale processing: High usage restrictions and throughput allow for large-scale processing.
• Cost Performance: Low cost while maintaining high performance.
• Real-time performance: Fast response speed, ideal for real-time processing.

9. Differences from o3

Features	OpenAI o3	OpenAI o4-mini
Positioning	High-performance inference model	High-speed, low-priced small model
Specialties	Complex analysis, advanced research	Mathematics, Coding, Visual Tasks
Suitable uses	Complex problem solving, advanced thinking support	Daily tasks, fast response, cost-consciousness
performance	The highest level	Very high in terms of size and cost ratio
Usage restrictions	Standard	Much higher than o3
cost	expensive	Cheap

10. Safety initiatives

OpenAI takes safety into consideration, reconstructs safety training data, improves rejection capabilities, and conducts strict evaluations.

11. summary

The o4-mini is a practical AI model that combines high speed, low cost and high performance. It is useful in situations where response speed, cost, and mass processing are important. Experience its performance with ChatGPT and APIs.

The above is the detailed content of [OpenAI] What is o4-mini? Explaining the main features, usage, and fee structure. For more information, please follow other related articles on the PHP Chinese website!