What is OpenAI o4-mini
OpenAI o4-mini is a small reasoning model launched by OpenAI, optimized for fast and cost-effective reasoning. OpenAI o4-mini performs excellently in mathematics, programming, and visual tasks, and is the best-performing model in the AIME 2024 and 2025 benchmarks. OpenAI o4-mini supports high-capacity, high-throughput reasoning tasks and is suitable for quickly processing large numbers of problems. OpenAI o4-mini has multimodal capabilities, integrates images into the chain of thought for reasoning, supports tool use, and can quickly generate detailed and thoughtful answers. Compared with previous-generation models, OpenAI o4-mini has significantly improved performance and cost-effectiveness. Currently, ChatGPT Plus, Pro, and Team users can see OpenAI o4-mini and OpenAI o4-mini-high in the model selector, replacing o1、o3‑mini and o3-mini-high. ChatGPT Enterprise and Edu users will gain access within a week. Developers are supported in using the model based on the Chat Completions API and Responses API.
Main features of OpenAI o4-mini
- Fast reasoning: Excels at quickly handling mathematics, programming, and visual tasks, suitable for high-throughput scenarios.
- Multimodal capabilities: Combines images and text for reasoning, supports image processing.
- Tool use: Calls tools such as web search and Python programming to assist in solving problems.
- High cost-effectiveness: Performance is better than the previous-generation o3-mini, with no price change, making it the first choice for upgrades.
- Safe and reliable: Trained for safety and supports refusing inappropriate requests.
Performance of OpenAI o4-mini
- Mathematical reasoning: In the AIME 2024 and 2025 benchmarks, OpenAI o4-mini achieves an accuracy of up to 93.4% without tools, and after integrating Python, its accuracy soars to 98.7%, close to a perfect score. In complex mathematical problem-solving ability, OpenAI o4-mini performs better than the previous-generation o3-mini. In some tasks, it is close to the full version of o3.
- Programming ability：
- SWE-Lancer: OpenAI o4-mini performs excellently, supports efficiently completing complex programming tasks, and delivers outstanding returns.
- SWE-Bench Verified (software engineering benchmark): OpenAI o4-mini performs excellently in tasks such as common algorithms, system design, and API calls, with both accuracy and efficiency higher than o3-mini.
- Aider Polyglot Code Editing (multilingual code editing benchmark): OpenAI o4-mini performs excellently in code editing tasks, including overall rewrites and patch-style modifications, with performance better than o3-mini in both.
- Multimodal capabilities：
- MMMU (university-level visual mathematics benchmark): OpenAI o4-mini supports combining images and mathematical symbols to solve problems, with an accuracy of 87.5%, far higher than the previous-generation o1's 71.8%.
- MathVista (visual mathematical reasoning): OpenAI o4-mini performs excellently in visual mathematical reasoning tasks such as geometric figures and function curves, with accuracy as high as 87.5%.
- CharXiv-Reasoning (scientific chart reasoning): OpenAI o4-mini can understand charts and diagrams in scientific papers, with an accuracy of 75.4%, significantly better than o1's 55.1%.
- Tool use：
- Scale MultiChallenge (multi-turn instruction following): OpenAI o4-mini supports handling complex multi-turn instruction tasks and correctly understands and executes multi-turn instructions.
- BrowseComp Agentic Browsing (browser tasks): Based on virtual browser search, clicks, page turning, and information integration, its performance is close to o3 and far exceeds traditional AI search capabilities.
- Tau-bench function calling: It performs stably in function calling tasks and supports accurate generation of structured API calls, but further optimization is needed in complex scenarios.
- Comprehensive testing：
- Expert-level comprehensive test (Humanity’s Last Exam): Without tools, the accuracy is 14.3%, and with plugins it increases to 17.7%, lower than o3's 24.9%, but it performs excellently among small models.
- Interdisciplinary PhD-level science questions (GPQA Diamond): The accuracy on science questions is 81.4%, slightly lower than o3's 83.3%, and it is already very outstanding among small models.
OpenAI o4-mini project address
- Project official website：https://openai.com/index/introducing-o4-mini/
Application scenarios of OpenAI o4-mini
- Educational tutoring: Helps students solve math and programming problems.
- Data analysis: Quickly generates data charts and analysis results.
- Software development: Generates code snippets and assists with code debugging.
- Content creation: Provides creative inspiration and combines image generation descriptions.
- Everyday queries: Answers questions based on search and image analysis.